You are on page 1of 215

Math 5120

Complex Analysis I
Spring 2009
Luke Rogers
Notes by Ben Salisbury
Last updated: May 5, 2009
Preface
This class is an introduction to complex analysis at the graduate level. A prac-
tical purpose of the class is to prepare students to take the qualifying exams.
Highlights of the course will be (not an exclusive list) analytic functions, mero-
morphic functions, the Cauchy Integral Formula, residues, maximum principle
and the Schwartz Lemma. For prelims, check out the Complex Analysis Study
Guide.
Notation
A(a, r, R) the annulus r < [z a[ < R
A(a, r, R) the closed annulus r [z a[ R
the boundary of
C the eld of complex numbers

C the Riemann sphere C


C() the continuous functions on
(z, z
1
, z
2
, z
3
) the cross ratio
D the open unit disk z : [z[ < 1
D the closed unit disk z : [z[ 1
D(a, R) the open disk z : [z a[ < R
D(a, R) the closed disk z : [z a[ R
T a family of functions
H the upper half-plane z : Im(z) > 0
Hol() the set of holomorphic functions on
Hol(, U) the set of holomorphic functions U

the path is homologous to

Im(z) the imaginary part of z = x +iy


L
1
(R) the integrable functions on R
the Laplacian operator
SL
2
(k) the group of 2 2 matrices with determinant 1 over k
R the eld of real numbers
Re(z) the real part of z = x +iy
Res(f, z) the residue of f at z
Z the ring of integers
N the monoid of natural numbers
z the complex conjugate of z C
CONTENTS
1 Complex Numbers 1
1.1 What is i? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 nth Roots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 The Riemann Sphere . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Complex Functions 9
2.1 Topology of the Complex Plane . . . . . . . . . . . . . . . . . . . 9
2.2 Dierentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Harmonic and Analytic Functions . . . . . . . . . . . . . . . . . . 12
2.4 Rational Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5 Partial Fractions . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.6 Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.7 The Exponential Function . . . . . . . . . . . . . . . . . . . . . . 24
3 Material for Later 29
3.1 Connectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Arc and Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4 Conformal Mappings 35
4.1 Linear Fractional Transformations . . . . . . . . . . . . . . . . . 35
4.2 Features of Mobius Transformations . . . . . . . . . . . . . . . . 38
4.3 Reection in a Circle . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.4 Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.5 Conformal Mappings of a Simply Connected Domain . . . . . . . 43
5 Integration Theory 57
5.1 Line Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.2 Applications of the Cauchy Integral Formula . . . . . . . . . . . 74
vii
viii CONTENTS
5.3 Taylor Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.4 Zeros of Analytic Functions . . . . . . . . . . . . . . . . . . . . . 80
5.5 Counting Zeros of Analytic Functions . . . . . . . . . . . . . . . 83
5.6 General Form of Cauchys Theorem . . . . . . . . . . . . . . . . 86
6 Residue Calculus 91
6.1 The Residue Theorem . . . . . . . . . . . . . . . . . . . . . . . . 91
6.2 Calculating Integrals . . . . . . . . . . . . . . . . . . . . . . . . . 94
7 Harmonic Functions 103
7.1 Harmonic Conjugates . . . . . . . . . . . . . . . . . . . . . . . . 103
7.2 Harmonic Functions and Normal Derivatives . . . . . . . . . . . 105
8 Series and Analytic Functions 113
8.1 Relationship Between Series and Analytic Functions . . . . . . . 113
8.2 Taylor Series Developments . . . . . . . . . . . . . . . . . . . . . 115
8.3 Laurent Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
9 Uniform Convergence 121
9.1 On Compact Subsets of . . . . . . . . . . . . . . . . . . . . . . 121
9.2 Normal Families . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
9.3 Riemann Mapping Theorem . . . . . . . . . . . . . . . . . . . . . 129
10 Selected Homework Solutions 131
10.1 Homework Set 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
10.2 Homework Set 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
10.3 Homework Set 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
10.4 Homework Set 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
10.5 Homework Set 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
10.6 Homework Set 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
10.7 Homework Set 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
10.8 Homework Set 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
10.9 Homework Set 9 . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
10.10Homework Set 10 . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
11 Preliminary Exams 191
11.1 Syllabus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
11.2 January 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
11.3 August 2008 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
11.4 August 2007 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
11.5 August 2006 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
11.6 August 2005 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
11.7 August 2004 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
11.8 August 2003 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
11.9 August 2002 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
11.10August 2001 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
CONTENTS ix
11.11August 2000 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
Bibliography 203
Index 203
CHAPTER 1
Complex Numbers
1.1 What is i?
We know i
2
= 1, but what is the motivation for this? One may suspect that
the reason was to solve quadratic equations. However, it was actually used to
study cubic equations.
We denote by C the set of complex numbers and write a typical element in
one of two ways: a = + i, which is consistent with [Ahl79], or z = x + iy,
which is consistent with most modern texts, where , , x, y R. If a = +i
and a

+i

, then a +b = ( +

) +i( +

) and
ab = ( +i)(

+i

) =

+i

+i

.
1.1 Example. Given only the information above, what is

i? From the mul-
tiplication formula above with a = +i, we have i = a
2
=
2

2
+2i, so
nding

i is equivalent to solving the equations 0 =
2

2
and 1 = 2. This
gives

i =

2
2
(1 + i). More generally [Ahl79, p. 3], for any complex number
a = +i, we have

a =
__
+

+
2
+i

[[
_
+

+
2
_
.
1
2 What is i?
1.2 Example. With a = +i, what is
1
a
? We have
1
a
=
1
+i
=
i

2
+
2
=

2
+
2
+
i

2
+
2
.
What these examples, along with the operations above, tell us is that C is a
eld. As an aside, note that C is the unique (up to isomorphism) eld containing
a root of a
2
+ 1 and containing R. To prove such an idea, one shows that the
operations on C are well-dened and do indeed form a eld. This shows existence
of a eld containing R and a root of a
2
+ 1. To show uniqueness, suppose i

is
another root of the equation and show R(i)

= C by the isomorphism i i

.
Geometrically speaking, the addition operation in C is the same as vector
addition in R
2
.
Figure 1.1: Addition in C is equivalent to vector addition in R
2
by viewing + i
as the point (, ).
Multiplication, however, is bit more interesting. Observe, as shown in Figure
1.2, that multiplication by i rotates the point by

2
radians in the counterclock-
wise direction.
So, multiplication by a complex number involves rotating (by the angle be-
tween the complex number and the real axis) and dilating (by the distance
Complex Numbers 3
Figure 1.2: This shows how multiplication in C is not exactly immediately
intuitive. Given a = + i, this shows how multiplication by i aects a.
between the origin and the complex number). This suggests we can write the
complex number +i by r(cos +i sin ), where r
2
=
2
+
2
.
1.3 Lemma. For r
1
, r
2
> 0 and
1
,
2
[0, 2), we have
(r
1
(cos
1
+i sin
1
))(r
2
(cos
2
+i sin
2
)) = r
1
r
2
(cos(
1
+
2
) +i sin(
1
+
2
)).
Proof. Follows from
cos(
1
+
2
) = cos
1
cos
2
sin
1
sin
2
sin(
1
+
2
) = cos
1
sin
2
+ sin
1
cos
2
and some algebra.
4 nth Roots
1.2 nth Roots
Let + i be a complex number with angle . Divide the angle into n pieces,
so we have n angles of size

n
. So the nth root of a = r(cos +i sin ) is
r
1/n
_
cos

n
+i sin

n
_
.
Because this is only a choice of an nth root, this particular root has a specic
name. Its called a primitive nth root. If we multiply a by any number
(cos + i sin ) such that
n
(cos(n) + i sin(n)) = 1, then we get another
nth root of a. So if = 1 and is such that n 0 mod 2, then we get our
primitive nth root.
We call the modulus of a complex number a = + i is [a[ =
_

2
+
2
,
and we dene the argument of a = +i to be arg a such that
a = [a[(cos(arg a) +i sin(arg a)).
The modulus [ [ has all the usual properties of an absolute value. Indeed, for
any a, b C, [a +b[ [a[ +[b[,

[a[ [b[

[a b[, and [ab[ = [a[[b[.


To this point, we have developed our theory based on the fact that i
2
= 1
and some algebraic denitions. But observe that i
2
= 1 also, so our theory
is symmetric under the reection i i. Insofar, we dene the complex
conjugate of a = +i as a = i. The complex conjugation map a a is
an idempotent ring homomorphism; that is, a +b = a + b, ab = a

b, and a = a
for all a, b C. Further, observe that
aa = ( +i)( i) =
2
+
2
= [a[
2
.
We already used this implicitly in the computation of the reciprocal of a.
Namely,
1/a =
a
[a[
2
.
For any a = +i C, we dene the real part and imaginary part of a
as Re(a) = and Im(a) = , respectively. Observe that
Re(a) =
a +a
2
and Im(a) =
a a
2i
.
Well use the real and imaginary parts to give a proof of the triangle in-
Complex Numbers 5
equality [Con73, p. 3]. For any a C, we have [a[ Re(a) [a[ and
[a[ Im(a) [a[. Hence Re(ab) [ab[ = [a[[b[. Therefore
[a +b[
2
= [a[
2
+ 2 Re(ab) +[b[
2
[a[
2
+ 2[a[[b[ +[b[
2
= ([a[ +[b[)
2
.
Taking square roots on both sides yields the triangle inequality.
Now write a = ab +b, for complex numbers a and b. Then [a[ = [ab +b[
and the triangle inequality implies [a[ [a b[ + [b[. Subtracting [b[ on both
sides yields [a[ [b[ [a b[. Since this inequality is symmetric in its roles of
a and b, we have shown

[a[ [b[

[a b[.
1.4 Example. For complex numbers a and b, consider the quantity
B :=

a b
1 ab

.
We intend to show B < 1 if [a[ < 1 or [b[ < 1, and B = 1 if [a[ = 1 or [b[ = 1.
Following [SL99, p. 7], it suces to assume a is real. Then the rst inequality
is reduced to proving (a b)(a b) < (1 ab)(1 ab).
Indeed, if we write a in its polar form a = r(cos + i sin ) and use Eulers
identity to write a = re
i
, we make the substitution b = be
i
and we nd

re
i
be
i
1 re
i
be
i

r b
1 rb

,
so we may assume that a is a real number 0 r 1. We want to show
(r b)(r b) (1 rb)(1 rb) (1.1)
with equality if and only if r = 1 or [b[ = 1. Expanding both sides, make the
necessary cancellations and move all the terms to the right to see that (1.1) is
equivalent to 0 (1 r
2
)(1 bb), so 0 (1 r
2
)(1 [b[
2
). Lastly, take the
derivative with respect to r to conclude the proof.
1.3 The Riemann Sphere
Thus far we have the complex place C. Lets add a single point at innity.
This is point, for example, to which all unbounded sequences tend. Topologically
(and informally) speaking, the plane with a point added looks like a sphere.
A common notation for C is

C.
6 The Riemann Sphere
More formally, we can identify

C with a sphere in three dimensions via
stereographic projection. This sphere is called the Riemann sphere. Lets
describe this so-called stereographic projection. In R
3
, identify C with the
rst two coordinates. Geometrically speaking, for every point z in C, the line
through z and (0, 0, 1) on the sphere intersects the sphere minus the north pole
at a unique point. Conversely, for each point (a, b, c) in the sphere dierent from
the north pole, the line through (0, 0, 1) and (a, b, c) intersects C at a unique
point. We use these maps to identify C, which equals sphere minus the pole,
with

C, which is the sphere, with corresponding to (0, 0, 1).
Figure 1.3: Stereographic projection identies a point in the complex plane C with
a unique point on the Riemann sphere
b
C.
Analytically, if z C is represented by (x, y, 0) in R
3
for x, y R, then the
line through z and (0, 0, 1) is (x, y, 0) +(1 )(0, 0, 1). This line intersects the
sphere when
2
(x
2
+y
2
) + (1 )
2
= 1; that is, = 0 or
=
2
x
2
+y
2
+ 1
.
When = 0, we dont get the intersection with the sphere for which we were
seeking. So the map C

C is given by
z
_
z +z
1 +[z[
2
,
z z
i(1 +[z[
2
)
,
[z[
2
1
1 +[z[
2
_
. (1.2)
Complex Numbers 7
In the other direction, the inverse map

C (0, 0, 1) C is given by
(a, b, c)
a +ib
1 c
. (1.3)
Under the Riemann sphere model, the unit disk in C corresponds to the bottom
hemisphere of

C. Moreover, the right half-plane in C (i.e., the complex numbers
with positive real part) corresponds to the the right hemisphere of

C. So there
is a correspondence between the unit disk and right half-plane in C simply by
rotating the Riemann sphere.
As an observation, lines in C are equivalent to circles through (0, 0, 1) on

C.
Additionally, circles in C correspond to circles on

C not containing (0, 0, 1).
What is the equation of a circle through (0, 0, 1) on

C. This circle lives in a
2-dimensional space, so the circle lies in a plane of the form
P = (a, b, c) R
3
: (, , ) ((a, b, c) (0, 0, 1)),
where (, , ) is a vector normal to the plane. In fact, the circle is the inter-
section of this plane P with

C. Now z C has image in

C given by (1.2). So if
z is on the curve corresponding to our circle, then
(, , )
_
z +z
1 +[z[
2
,
z z
i(1 +[z[
2
)
,
2
1 +[z[
2
_
= 0.
So
(z +z) i(z z) 2 = 0
and it can be veried that this is (a form of) the general equation of a line in C
(see exercises).
One can put a metric on

C to C via the stereographic projection, at which
point we nd if (a, b, c) corresponds to z and (a

, b

, c

) corresponds to z

, then
dist((a, b, c), (a

, b

, c

)) =
2[z z

[
_
(1 +[z[
2
)(1 +[z

[
2
)
and
dist((a, b, c), (0, 0, 1)) =
2
_
1 +[z[
2
.
These formulae dene the spherical metric on C.
CHAPTER 2
Complex Functions
In studying complex functions, one must rst distinguish which class of functions
one wishes to perform analysis; that is to say, functions are are suitably nice:
continuous, dierentiable, etc. Before we can discuss such properties, we must
rst discuss the topology on C. Indeed, we use the usual Euclidean distance on
R
2
.
2.1 Topology of the Complex Plane
2.1 Denition. We have
lim
za
f(z) = A
if and only if for all > 0 there exists > 0 such that [f(z) A[ < on
[z a[ < .
Observe since the metric on C is the same as the usual distance on R
2
, this
is the usual notion of a limit in R
2
. In particular, f(z) A as z a in C if
and only if Re(f(z)) Re(A) and Im(f(z)) Im(A) as z a.
There are also such things as
lim
z
f(z) = A, lim
za
f(z) = , lim
z
f(z) = .
It is left to reader to develop appropriate denitions for each of these limits.
9
10 Differentiation
2.2 Proposition. The usual limit laws from R are valid in C. Moreover, if
f(z) A as z a, then [f(z)[ [A[ as z a. Lastly, if f(z) A as z a,
then f(z) A.
The usual notion of continuity also carries over to complex functions.
2.3 Denition. A function f is continuous at a point a if f(z) f(a) as
z a. A function f is continuous on a set if it is continuous at each point in
the set. The phrase f is continuous means f is continuous on its domain.
As a consequence of Proposition 2.2, we have the following theorem.
2.4 Theorem. Continuity of complex functions is preserved by sums, prod-
ucts, quotients (on the domain of the quotient), complex conjugation, real part,
imaginary part, and modulus.
2.5 Example. From [SL99, p. 10]. Consider the function
f(z) = lim
n
z
n
1
z
n
+ 1
, [z[ , = 1.
We show that this limit exists but it is not possible to dene f(z) when [z[ = 1
in such a way that makes f continuous. Suppose rst that [z[ < 1. Then

z
n
1
z
n
+ 1
(1)

2z
n
z
n
+ 1

0
as n , so f(z) = 1. If [z[ > 1, then

z
n
1
z
n
+ 1
1

=
2
[z
n
+ 1[
0
as n , so f(z) = 1. From these results we see that we cannot dene f(z)
when [z[ = 1 so as to make f continuous.
2.2 Dierentiation
2.6 Denition. Let f be a complex function. We dene the derivative of f at
a as
f

(a) = lim
za
f(z) f(a)
z a
= lim
0
f(a +) f(a)

.
Complex Functions 11
2.7 Example. Its computation time! Let z = x + iy and f(z) = u(x, y) +
iv(x, y), where both u and v are maps R
2
R. Compute f

(0) by considering
the limit where 0 for R, and where 0 for purely imaginary.
Indeed, if R, then
f

(0) = lim
0
(u(, 0) +iv(, 0)) (u(0, 0) +iv(0, 0))

=
u
x
+i
v
y
and for imaginary
f

(0) = lim
h0
(u(0, h) +iv(0, h)) (u(0, 0) +iv(0, 0))
ih
=
1
i
_
u
x
+i
v
y
_
.
2.8 Theorem. If f

(a) exists then


u
x

z=a
=
v
y

z=a
and
v
x

z=a
=
u
y

z=a
.
These are called the Cauchy-Riemann equations.
If we think of f as representing a map R
2
R
2
by (x, y) (u, v), then
the derivative of f as a map R
2
R
2
is represented by a 2 2 matrix
Df =
_
u/x u/y
v/x v/y
_
.
By the Cauchy-Riemann equations, the row vectors in
_
u/x u/y
v/y v/x
_
.
are orthogonal and have the same length. Thus
Df =
_
_
u
x
_
2
+
_
u
y
_
2
_
O,
where O is an orthogonal matrix. In fact, O is an element of SO
2
(R), so it
represents a rotation. Observe also that this linear mapping scales area by a
factor of
_
u
x
_
2
+
_
u
y
_
2
= Jacobian of f as a map R
2
R
2
.
12 Harmonic and Analytic Functions
Said dierently, suppose g : R
2
R
2
and g is dierentiable as a map of
this type. That is to say, there is a linearization of g at every point in R
2
. It
is reasonable to think that a disk around any point in R
2
gets maps to another
ellipse in R
2
under g. The ellipse may be rotated and may be stretched, but
the ellipse is not going to be deformed into a line (Figure 2.1). Geometrically
speaking in C, innitesimal disks in C get mapped to rotations and dilations of
the original disk under dierentiable maps C C.
Figure 2.1: A dierentiable map R
2
R
2
is equivalent to mapping every disk in
R
2
to some scaling and rotation of the original disk. However, unlike complex
dierentiable functions C C, it does not preserve angles.
This relates to the notion of a conformal (angle-preserving) map. In R
2
,
a dierentiable map may not preserve the angle between tangent vectors of
two curves meeting at a given point. However, complex dierentiable functions
preserve angles.
2.3 Harmonic and Analytic Functions
Suppose f(z) = u + iv, where u(x, y) and v(x, y) are C
2
, and f is complex
dierentiable. The Cauchy-Riemann equations imply

2
u
x
2
=

2
v
xy
and

2
u
y
2
=

2
v
xy
.
Complex Functions 13
Thus _

2
x
2
+

2
y
2
_
u = 0 and
_

2
x
2
+

2
y
2
_
v = 0.
2.9 Denition. The dierential operator
:=
_

2
x
2
+

2
y
2
_
is called the Laplacian. Any C
2
function with vanishing Laplacian is called
harmonic.
2.10 Theorem. If f is suciently smooth and is analytic, then Re(f) = u and
Im(f) = v are harmonic functions.
The converse is not true. If u and v are harmonic functions such that the
function f = u+iv is analytic, then u are v are called harmonic conjugates.
2.11 Theorem. If u and v are C
1
functions R
2
R, where we may possibly
restrict the function to an open set U R
2
, such that the Cauchy-Riemann
equations hold, then f = u +iv is analytic.
Proof. Fix a point = (a, b) R
2
and write z = x +iy. The linearization of u
at (a, b) is
u(x, y) = u(a, b) +u
x
(a, b)(x a) +u
y
(a, b)(y b) +o([x a[ +[y b[),
and similarly for v(x, y). Then
f(z) f()
z
u
x
(a, b) +iu
y
(a, b)
=
(u(x, y) u(a, b)) +i(v(x, y) v(a, b))
(x a) +i(y b)
u
x
(a, b)(x a) iv
y
(a, b)(y b) u
y
(a, b)(y b) iv
x
(a, b)(x a)
(x a) +i(y b)
=
o([x a[ +[y b[)
(x a) +i(y b)
0 as z .
Now some dodgy formal reasoning that gives intuition. Suppose we have a
function f : R
2
R
2
by f(x, y) = (u(x, y), v(x, y)). Think of x =
1
2
(z + z)
and y =
1
2i
(z z). So we can express f as f(z, z). Formally,

z
=
1
2
_

x
i

y
_
and

z
=
1
2
_

x
+i

y
_
.
14 Harmonic and Analytic Functions
Notice the Cauchy-Riemann equations say f/z = 0. This says that a complex
dierentiable functions are a function of z, not of both z and z.
2.12 Example. If f and g are both analytic functions, we show gf is analytic
and satises the usual chain rule [Con73, p. 34]. Let G be the domain on which
f is dened and the domain of g. Fix z
0
in G and choose r > 0 such that
B(z
0
, r) G. We must show that if 0 < [h
n
[ < r and h
n
0 as n , then
lim
n
g(f(z
0
+h
n
)) g(f(z
0
))
h
n
exists and equals g

(f(z
0
))f

(z
0
). (Why is this sucient?)
Case 1. Suppose f(z
0
) ,= f(z
0
+h
n
) for all n. Then
g(f(z
0
+h
n
)) g(f(z
0
))
h
n
=
g(f(z
0
+h
n
)) g(f(z
0
))
f(z
0
+h
n
) f(z
0
)

f(z
0
+h
n
) f(z
0
)
h
n
.
Since f is analytic,
lim
n
f(z
0
+h
n
) f(z
0
) = 0,
so we have
lim
n
g(f(z
0
+h
n
)) g(f(z
0
))
h
n
= g

(f(z
0
))f

(z
0
).
Case 2. Now suppose f(z
0
) = f(z
0
+h
n
) for innitely many n. Write h
n

as the union of two sequences k


n
and
n
where f(z
0
) ,= f(z
0
+ k
n
) and
f(z
0
) = f(z
0
+
n
) for all n. Since f is analytic,
f

(z
0
) = lim
n
f(z
0
+
n
) f(z
0
)

n
= 0.
Also,
lim
n
g(f(z
0
+
n
)) g(f(z
0
))

n
= 0.
So by Case 1,
lim
n
g(f(z
0
+k
n
)) g(f(z
0
))
k
n
= g

(f(z
0
))f

(z
0
) = 0.
Therefore
lim
n
g(f(z
0
+h
n
)) g(f(z
0
))
h
n
= 0 = g

(f(z
0
))f

(z
0
).
Complex Functions 15
The general case follows from these two.
There is a formal method for obtaining the harmonic conjugate of a harmonic
function. That is, given u C
2
with u = 0. We want to nd v such that u+iv
is analytic. One solution, from multivariable calculus, is familiar. Compute the
partial derivative u
x
, we have v
y
= u
x
, so
v =
_
u
x
dy +g(x).
Then compute v
x
by
v
x
=

x
_
u
x
dy +g

(x).
Set this equal to u
y
. This gives an equation for g

(x), then nd g(x) and


substitute.
Here is another solution. Write
f(z) = 2u
_
z
2
,
z
2i
_
u(0, 0),
which is valid when u is a polynomial or rational function.
Lets return to analytic functions. Some common examples of analytic func-
tions include the constant functions and the function f(z) = z. The product
and sum of analytic functions are analytic. The proofs of the latter are identical
to the real variable case, just switch x with z.
2.13 Theorem. The polynomial functions are analytic.
Any polynomial P(z) C[z] is of the form a
0
+a
1
z + +a
n1
z
n1
+a
n
z
n
,
where a
n
,= 0. By denition, the degree of P is n with the convention that if
P(z) = 0, then deg(P) = 0. A fact (that will be proved later) is the fundamental
theorem of algebra, which says any polynomial of degree at least 1 has at least
one root. We know from algebra, if P(a) = 0, then P(z) = (z a)Q(z) where
deg(Q(z)) = deg(P(z)) 1. By induction we have the following.
2.14 Proposition. Any polynomial P(z) splits in C. That is, there exist
j
,
for j = 1, . . . , deg(P), such that
P(z) = a
n
n

j=1
(z
j
),
where n = deg(P).
16 Rational Functions
In the above, the
j
need not be distinct. If some root is repeated exactly
k times, so P(z) = (z )
k
Q(z) and Q() ,= 0, then P(z) has a zero of order
k at . Observe,

zeros
order of the zero = deg(P).
2.15 Lucas Theorem. Suppose P(z) is a polynomial. If is a half-plane
that contains no roots of P(z), then no roots of P

(z) are in .
2.16 Corollary. The zeros of P

(z) lie in the convex hull of the zeros of P(z).


Observe that the line containing two zeros of P(z) forms a half-plane. One
side of the half-plane contains no zeros of P(z), so by Lucas theorem that half-
plane contains no zeros of P

(z). So since the convex hull of the zeros is the


intersection of all the half-planes containing the zeros of P, we have the zeros
of P

(z) are in the convex hull of the zeros of P(z).


Proof of Lucas theorem. Look at
F(z) :=
P

(z)
P(z)
=
n

j=1
1
z
j
.
Let be a half-plane containing no zeros of P(z). Then
F(c(z b)) =
n

j=1
1
c(z
j
b)
has no zeros in the right half-plane, where b and c are the complex numbers
translating the half-plane into H := z : Re(z) > 0. Since F(c(z b)) has
no zeros in H, Re(z
j
b) > 0 for z . Hence Re(
1
zjb
) > 0, so
Re
_
n

j=1
1
z
j
b
_
> 0
as needed.
2.4 Rational Functions
Let R(z) = P(z)/Q(z) for P(z) and Q(z) in C[z] such that P(z) and Q(z)
have no common zeros. Such a function R(z) is called a rational function. We
consider these functions on

C because the theory is nicer there.
Complex Functions 17
Any rational function R(z) has zeros exactly at the zeros of P(z), and the
order of each zero is the same as for P(z). In particular, if n = deg(P(z)), then
R has n zeros in C counting multiplicities.
2.17 Denition. A zero of Q(z) is called a pole of R(z).
Like with P(z), the number of poles of R(z) is equal to the number of zeros
of Q(z) in C, namely R(z) has m = deg(Q(z)) poles in C. However, we have
not yet considered the case when R is near innity. Write
R(z) =
a
0
+a
1
z + +a
n
z
n
b
0
+b
1
z + +b
m
z
m
=
z
n
(

n
j=0
a
j
z
jn
)
z
m
(

m
j=0
b
j
z
jm
)
.
If n > m, then R(z) as z . In which case, we say R(z) has a pole at
innity of order n m . If n = m, then R(z) a
n
/a
m
as z . If n < m,
then R(z) 0 as z and the order of the zero is m n. A justication
of this is that since every point on

C is the same, if we reect

C such that
0, which is done for example by z 1/z, then R(1/z) near z = 0 has the
same behavior as R(z) at . It is an easy exercise to show
R(z) has
_

_
pole of order n m, n > m
zero of order mn, n < m
neither, n = m.
2.18 Example. Let
R(z) =
(z 2)
2
(z
3
+ 1)
z
4
+z
2
+ 1
.
The numerator has at least ve zeros, so R(z) has ve zeros in C. The func-
tion R(z) has four poles in C and one pole at innity. So in general, R(z) =
P(z)/Q(z), with n = deg(P(z)) and m = deg(Q(z)), has n zeros in C, m poles
in C, so mn zeros if m > n and n m poles if n > m. Hence
#(zeros of R(z)) =
_
_
_
n, n m
m, n < m
= max(m, n)
and
#(poles of R(z)) =
_
_
_
m, m n
n, m < n
= max(m, n).
18 Partial Fractions
2.19 Theorem. If R(z) = P(z)/Q(z) is a rational function such that P(z) and
Q(z) have no common zeros, then the number of zeros in

C equals the number
of poles in

C, and both values are max(deg(P(z)), deg(Q(z))).
2.5 Partial Fractions
Recall, for example,
1
z
2
1
=
A
z 1
+
B
z + 1
=
from calculus. How does one prove this always possible for a rational function
R(z)?
2.20 Theorem. A rational function can always be written in partial fraction
form.
Proof. Consider the case where R(z) has a pole at innity. That is, R(z) =
P(z)/Q(z) where deg(P(z)) > deg(Q(z)). Then (by the Euclidean algorithm)
R(z) = G(z) +
H(z)
Q(z)
, deg(H(z)) = deg(Q(z)).
Call G(z) the singular part, or principal part, of R at .
Similarly, if C is such that Q() = 0 (i.e., R has a pole at ), then we
could look at H(z)/Q(z). Consider
H( +
1
z
)
Q( +
1
z
)
,
which has a pole at . So the same argument as the last paragraph implies
H( +
1
z
)
Q( +
1
z
)
= G

(z) +
J(z)
Q(z)
.
Then
H(z)
Q(z)
= G

_
1
z
_
+
J(
1
z
)
Q(
1
z
)
has neither pole nor zero at . Call G

the singular part at . By a nite


number of operations,
R(z) G(z)

j
G
j
_
1
z
j
_
,
Complex Functions 19
such that
j
C are the roots of Q in C, and thus has no poles in

C, so it is
constant.
So not only did we prove the theorem, but we obtained an algorithm for
nding the partial fractions decomposition. We have
R(z) = c +G(z) +

j
G
j
_
1
z
j
_
,
where c C is a constant.
2.6 Power Series
A power series is a series of the form

j0
a
j
(z z
0
)
j
,
where z
0
is called the center. By translation, it suces to assume that z
0
= 0.
2.21 Example. The fundamental example of a power series is the geometric
series
a

j0
(bz)
j
.
This series converges to
a
1 bz
for [bz[ < 1 and diverges for [bz[ 1. Clearly it diverges for [bz[ > 1 since
bz ,0. To see that it converges for [bz[ < 1, look at
(1 bz)(1 +bz + (bz)
2
+ + (bz)
n1
) = 1 (bz)
n
.
This implies
n1

j=0
(bz)
j
=
1 (bz)
n
1 bz
, bz ,= 1.
Letting both sides approach innity yields
n1

j=0
(bz)
j
=
1 (bz)
n
1 bz

1
1 bz
.
20 Power Series
2.22 Theorem. Let

j0
a
j
z
j
be a power series.
(a) There is a radius of convergence R, with 0 R , such that the series
converges absolutely for [z[ < R and the series diverges for [z[ > R.
(b) For 0 < R, the series converges uniformly on [z[ . Said dierently,
for every > 0 there exists a positive integer N such that
sup
|z|

jN
a
j
z
j

< .
(c) The function
f(z) =

j0
a
j
z
j
is analytic on [z[ < R and
f

(z) =

j0
ja
j
z
j1
with the same radius of convergence R as the power series expansion of f.
(d) The radius of convergence is dened by
1
R
= limsup
n
[a
n
[
1/n
,
where we use the convention that 1/= 0 and 1/0 = .
Proof. The proofs of (a) and (b) follow from [Con73, p. 31]. Let R be dened as
in (d). If [z[ < R there is an r with [z[ < r < R. Thus there exists an integer N
such that
n
_
[a
n
[ < r
1
for all n N (since r
1
> R
1
). But then [a
n
[ < r
n
and so [a
n
z
n
[ < [z[
n
r
n
for all n N. This says that the tail end

nN
a
n
z
n
is dominated by the series

n0
[z[
n
r
n
, and since [z[/r < 1, the power series
converges absolutely for each [z[ < R.
Now suppose r < R and choose such that r < < R. As above, let N be
such that [a
n
[ <
n
for all n N. Then if [z[ r, we have [a
n
z
n
[ r
n

n
and
r/ < 1. Hence the Weierstrass M-test gives that the power series converges
uniformly on z : [z[ r.
It remains to show that the series diverges for [z[ > R. Let [z[ > R and
choose r with [z[ > r > R. Hence r
1
< R
1
. By the denition of limsup,
this gives innitely many integers n with r
1
<
n
_
[a
n
[. It follows that [a
n
z
n
[ >
Complex Functions 21
[z[
n
r
n
and, since [z[/r > 1, these terms become unbounded. Hence the series
diverges.
Last we need to show that one can dierentiate a power series term-by-
term [Rud87, p. 199]. Indeed, if the original series converges in D(0, R), then
the root test shows that the series of derivatives converges there, also. Denote
the sum of the series of derivatives by g(z) and x w D(0, R). Choose so
that [w[ < < R. If z ,= w, we have
f(z) f(w)
z w
g(w) =

n1
a
n
_
z
n
w
n
z w
nw
n1
_
.
The expression in parentheses is 0 if n = 1, and it is
(z w)
n1

k=1
kw
k1
z
nk1
(2.1)
if n 2. If [z[ < , the absolute value of the sum in (2.1) is less than
n(n 1)
2

n2
,
so

f(z) f(w)
z w
g(w)

[z w[

n2
n
2
[a
n
[
n2
. (2.2)
Since < R, the last series converges. Hence the left side of (2.2) tends to 0 as
z w. This says that f

(w) = g(w), and completes the proof.


As a consequence of this theorem (that sometimes people miss), if

j0
a
j
z
j
converges at a point w, then it converges absolutely on [z[ < [w[.
2.23 Example. Can you nd the power series expansion of a function? Suppose
f(z) =
2z
3+z
. We will use the geometric series. Indeed,
2z
3 +z
= 2z
1
3(1
z
3
)
=
2z
3

j0
_
z
3
_
j
.
Another example,
1
(1 z)
2
=

j0
jz
j1
by dierentiation the geometric series term-by-term. So we can get the power
22 Power Series
series expansion of any rational function by using partial fractions (and some
combination of algebraic manipulation and dierentiation/integration) and the
geometric series.
2.24 Example. From [SL99, p. 11]. We discuss the convergence of the series

n1
z
n1
(1 z
n
)(1 z
n+1
)
.
Let
U
n
=
z
n
(1 z
n
)(1 z
n+1
)
and let D(z) = (1 z
n
)(1 z
n+1
). Then
U
n
=
z
n
(1 z)
D(z)(1 z)
=
1
1 z
_
z
n
z
n+1
D(z)
_
=
1
1 z
_
(z
n
1) + (1 z
n+1
)
(1 z
n
)(1 z
n+1
)
_
=
1
1 z
_

1
1 z
n+1
+
1
1 z
n
_
,
where the last step is found using partial fractions. We get a telescoping sum,
whence
n

k=1
U
k
=
1
1 z
_
1
1 z

1
1 z
n+1
_
and therefore
S
n
(z) =
n

k=1
z
k1
(1 z
k
)(1 z
k+1
)
=
1
z(1 z)
_
1
1 z

1
1 z
n+1
_
.
If [z[ < 1, then
1
1 z
n+1
1, n ,
and therefore
S
n
(z)
1
(1 z)
2
, n .
If [z[ > 1, then
1
1 z
n+1
0, n ,
so
S
n
(z)
1
z(1 z)
2
.
Complex Functions 23
Now consider this series on the compact set [z[ K < 1. A little algebra
combined with the calculations above show

S
n
(z)
1
(1 z)
2

1
z(1 z)

z
n+1
1 z
n+1

.
But [1 z
n+1
[ 1 [z[
n+1
1 K
n+1
, so we get the estimate

S
n
(z)
1
(1 z)
2

1
1 K
K
n
1 K
n+1
for all z in the region [z[ K < 1. Now K
n
0 so the convergence is uniform
in the given region.
If [z[ M > 1, then

S
n
(z)
1
z(1 z)
2

1
z(1 z)

1
1 z
n+1

1
M(M 1)
1
M
n+1
1
and M
n+1
, so the convergence is uniform in the region [z[ M > 1.
What happens on the boundary of the disk of convergence? Anything, more
or less. For example, the series

n0
nz
n
diverges on [z[ = 1, but

n0
z
n
n
2
converges on [z[ = 1. What about

n0
z
n
? It converges on [z[ = 1, except at
z = 1. (This series is the source of the Abel summability theory.)
There is a method for dealing with things like

n0
z
n
due to Abel. The
starting point is partial summation; i.e., summation by parts (equivalent to
integration by parts).
2.25 Abels Theorem. If A
n
= a
0
+ +a
n
and B
n
= b
0
+ +b
n
, then
n

j=0
a
j
b
j
= a
n
B
n

j=0
(a
j+1
a
j
)B
j
.
This has an application to power series.
2.26 Abels Theorem. Suppose

n0
a
n
converges. Then
lim
z1

n0
a
n
z
n
=

n0
a
n
.
Proof. We follow [Ahl79, p. 41]. We may assume

j0
a
j
= 0, for this can be
obtained by adding a constant to a
0
. We write s
n
= a
0
+a
1
+ +a
n
and make
24 The Exponential Function
use of summation by parts. Indeed,
s
n
(z) = a
0
+a
1
z + +a
n
z
n
= s
0
+ (s
1
s
0
)z + + (s
n
s
n1
)z
n
= s
0
(1 z) +s
1
(z z
2
) + +s
n1
(z
n1
z
n
) +s
n
z
n
= (1 z)(s
0
+s
1
z + +s
n1
z
n1
) +s
n
z
n
.
But s
n
z
n
0, so we obtain the representation
f(z) = (1 z)

n0
s
n
z
n
.
We are assuming that [1 z[ K(1 [z[), say, and that s
n
0. Chose N
so large that [s
n
[ < for n N. The remainder of the series

n0
s
n
z
n
, from
n = N on, is then dominated by the geometric series

nN
[z[
n
=
[z[
N
1 [z[
<

1 [z[
.
It follows that
[f(z)[ [1 z[

m1

k=0
s
k
z
k

+K.
The rst term on the right can be made arbitrarily small by choosing z su-
ciently close to 1, and we conclude that f(z) 0 when z 1 subjected to the
stated restriction.
2.7 The Exponential Function
We dene
exp(z) = e
z
=

n0
z
n
n!
.
Observe that limsup
n

n! = , so e
x
is an analytic function on all of C. Then
d
dz
e
z
=

n0
nz
n1
n!
=

n1
z
n1
(n 1)!
= e
z
.
Complex Functions 25
The dierential equation implies
d
dz
(e
z
e
cz
) = 0,
for any constant c.
2.27 Lemma. If f is an analytic function such that f

0, then f is constant.
Proof. Use the same fact for real variable functions on real and imaginary parts
in real and imaginary directions. So for the sake of completeness, we include
the real variable proof here, from [Lan97, p. 72]. Consider f such that f is
continuous on [a, b] and dierentiable on (a, b). Then by the mean value theorem,
for any x (a, b) there exists c (a, x) such that
f(x) f(a) = f

(c)(x a) = 0.
Hence f(x) = f(a), and f is constant.
Hence e
z
e
cz
is constant by the lemma, and, in fact e
z
e
cz
= e
c
, which is
obtained from e
z
= 1 by substituting in the power series. The conclusion is
that e
a+b
= e
a
e
b
; i.e., exp: C C

is a group homomorphism. Combining


this result with the observation that e
z
= e
z
(because coecients in the power
series expansion are real), we get [e
iy
[
2
= e
iy
e
iy
= 1.
If we dene
cos z =

n0
(1)
n
z
2n
(2n)!
, sin z =

n0
(1)
n
z
2n+1
(2n + 1)!
,
we discover these also have radius of convergence , so they are analytic on all
of C. Algebra allows us to compute
e
x+iy
= e
x
cos y +ie
x
sin y = e
x
(cos y +i sin y),
and
cos z =
e
iz
+e
iz
2
, sin z =
e
iz
e
iz
2i
.
We can show that e
z
is periodic. Indeed, if there exists c such that e
c
= 1,
then e
z+c
= e
z
e
c
= e
z
, and inductively e
z+kc
= e
z
for all k Z. Since e
x+iy
=
e
x
e
iy
and [e
iy
[ = 1 we have [e
x+iy
[ = [e
x
[ = e
x
= 1 if and only if x = 0, so it
26 The Exponential Function
suces to assume c is imaginary to nd c such that e
c
= 1. So we need
1 = e
iy
= cos y +i sin y,
so we need sin y = 0 and cos y = 1. There is a detailed proof of the existence
of a such a y in [Ahl79]. It uses only cos
2
y + sin
2
y, the intermediate value
theorem, and the mean value theorem. The conclusion is there is such a y. We
call this smallest such value 2 and have proved as result the following.
2.28 Theorem. The exponential function is 2i-periodic (and has no smaller
period), and both cosine and sine are 2-periodic.
Figure 2.2: Writing e
z
= e
x
(cos y + i sin y) and using the properties of exp on R
and that |e
iy
| = 1, horizontal lines i in C get mapped to rays emanating from the
origin with argument in C. Vertical lines c + iy get wrapped around a circle. In
other words, the exponential map converts the rectangular grid to a polar grid.
Observe that [e
x+iy
[ = e
x
and arg(e
x+iy
) = y mod 2Z.
Next well talk about the logarithm. The logarithm in C is a multiple valued
function. Write w = e
z
= e
x+iy
. We say z is a logarithm of w and z = x +iy
where x is determined from [w[ = e
x
, so x = log [w[ (the real logarithm), and
y is arg(w). Hence log w = log [w[ + i arg(w), with the convention that log z is
the single-valued real logarithm if z is real and log z is the multi-valued complex
logarithm if z is complex. The reason we use this convention stems from the
fact that we can use exp and log to dene w
z
= exp(z log w). Observe that the
multi-valued part of this function occurs since arg(w) is a set that is really
only well-dened modulo 2Z.
So we can write any complex number in polar form as re
i
. Writing a com-
Complex Functions 27
plex number in this form reinforces the fact that multiplying complex numbers
results in rotation by their arguments; i.e., r
1
e
i1
r
2
e
i2
= r
1
r
2
e
i(1+2)
.
Another important example of a power series that denes an analytic func-
tion is the following.
2.29 Theorem. Let (X, ) be a probability space and let : X C be inte-
grable. Then
_
X
d(x)
(x) z
is analytic on C Int((X)).
2.30 Example. Take a rectiable curve in C and call it X, with measure equal
to the normalized length measure. Let be the inclusion map X C. Then
f(z) =
_
X
d(x)
x z
denes an analytic function C X.
This is a prototypical example of a singular integral, and studying when
x z is an example of singular operators, which is an important subject in
harmonic analysis.
Proof. Let a C Int((X)). Write
1
(x) z
=
1
(x) a (z a)
=
1
(x)

n0
_
z a
(x) a
_
n
,
which converges since it is dominated by a geometric series which converges on
the disk [z a[ < dist(a, (x)). Then
_
X
d(x)
(x) z
=
_
X

n0
(z a)
n
((x) a)
n+1
d.
The convergence is uniform on X if [z a[ r < dist(a, (X)). So on a such a
disk,

n0
__
X
d(x)
((x) a)
n+1
_
(z a)
n
,
which is a power series.
CHAPTER 3
Stu we will need later. . .
3.1 Connectedness
The reader is expected to be familiar with open and closed sets, completeness,
and compactness as well as its equivalent characterizations. We focus primarily
on the relative topology and connectedness.
Let X be a topological space and Y X. Recall (or dene) the relative
topology on Y by U Y is open if U = V Y for some open set V in X.
Beware! If Y has its own intrinsic topology, the intrinsic topology and the
relative topology induced from X may not coincide.
3.1 Example. Consider the interval [a, b] as a subset of R
2
. This set is closed
with its intrinsic topology. However, this interval is open in the relative topology
since it is the subset of some open disk in the plane.
We are now ready to dene connectedness. Let X be a topological space.
We dene a separation of X to be a pair of nonempty open sets such that
X = A B and A B = . The space X is said to be connected if there is
no separation of X.
This particular denition may seem a bit cumbersome. Usually the way this
is used is in one of the following forms.
If A X is both open and closed, then A = X or A = . This statement
29
30 Connectedness
implies that X is connected. Indeed, if A is open and B is open, then A
is also closed since A = X B.
If we know X is connected and we have a nonempty subset A and A has
some property, we show A = X by showing A is both open and closed.
3.2 Example. On the real line R, any interval [a, b], (a, b], (a, b), [a, b) are
examples of connected sets, where a < b. Indeed, suppose [a, b] = A B where
A and B are nonempty, disjoint, closed and open subsets of [a, b]. Let c = sup A.
If c [a, b), then c A because A is closed. But A is open, so (c , c + )
is a neighborhood of c that is contained in A. So c + A, which contradicts
c = sup A. Hence sup A = b and b A. The same argument shows b B, hence
b A B and [a, b] is connected. The proof that (a, b] is connected follows
exactly as above, and [a, b) uses the inmum rather than the supremum.
How does one show (a, b) is connected? A possible argument is to start by
dividing the interval in half. At the place where the division is made, we have
sup A = inf B, so no separation exists. This way also works for the intervals
above. So we are looking for the point where A and B meet. That is, a
point x such that every neighborhood of x intersects both A and B. Do this by
subdividing (a, b) into left and right halves. This gives the following possibilities:
either all of A is on one side and all of B on the other (in which case we can stop
and let x be the midpoint), or one side contains points from both A and B. In
this case, keep this half and throw away the other and set x
1
= (a +b)/2. Now
if we cannot stop, repeat at a smaller scale. By induction, either we stopped
from nding our desired x or we built a sequence x
n
that is Cauchy (because
[x
j+1
x
j
[ 2
j
[a b[). This implies
[x
k
x
j
[ [a b[
k

=0
2

[a b[2
j+1
.
This has a limit, call it x. Now any neighborhood of x contains an interval
from the construction, so it contains points of both A and B. Now we have the
desired x and every (deleted) neighborhood contains points of A and of B, so x
is a limit point of both A and B. But A and B are both closed, so x A and
x B. Hence A B is nonempty which says no separation of (a, b) exists.
3.3 Theorem. The connected sets in R are precisely the intervals and the empty
set.
Material for Later 31
Proof. We have shown that the intervals are connected, so it remains to show
if a subset of R is connected then it must be an interval. Let X R be a
connected set. Let a = inf X and b = sup X, with the possibility that either
value could be . If a > b, then X = . If a = b, then X is a point, so it is
the interval [a, a] = [b, b]. Now suppose a < b. If there exists x (a, b) such that
x / X, then [a, x) X = A and (x, b] X = B, which are both open and closed
subsets of X such that A B = and A B = X. Hence X is disconnected,
which is a contradiction. So x X and (a, b) X. Thus X is one of (a, b),
(a, b], [a, b), or [a, b].
3.4 Theorem. Let X be connected and let f : X Y be a continuous map.
Then f(X) is connected.
Proof. The preimage of a separation of f(X) is a separation of X.
As a corollary, we have the intermediate value theorem.
3.5 Theorem. A subset X of R
2
or C is open and connected if and only if any
two points in X can be joined by a polygonal path. In fact, such a path can be
chosen to have all edges parallel to the axes.
Proof. (=) Let a X. The set of all points connected to a by a polygonal
path is open, because if there exists a polygon from a to b X and D is a
disk containing b and D is contained in X, then there exists a polygonal path
from b to any point in D. This implies D is contained in the set of points
polygonally connected to a. Also, the set of points not polygonally connected
to a is open. This is the same argument. Let A be the set of points connected
to a by a polygonal path and let B be the set of points not connected to a by
a polygonal path. Then A B = X, A B = , and A and B are open. Since
B is connected, B is empty.
(=) Suppose there is a separation of X, namely (A, B). Take a point a A
and b B. The proof is left to the reader, but it appeals to the fact that the
continuous image of a connected set is connected.
We can dene connectedness in a more local form. If A is a set, a connected
component is a maximal connected subset under inclusion.
3.6 Theorem. Any set can be decomposed into disjoint connected components.
32 Connectedness
Proof. Let A be a set. If A = , then there is nothing to show. If A is nonempty,
take a A and let ( = B a : B connected, B A. Then a (, so ( is
nonempty. Let
C
a
=
_
BC
B.
We wish to show that C
a
is connected, so suppose to the contrary that C
a
is
not connected. Let (D, E) be a separation of C
a
. Without loss of generality,
take a D. Then there exists e E, since E is nonempty, so e B for some
B C
a
. In particular, the pair (DB, EB) forms a separation of B, which is
a contradiction with the connectivity of B. Hence C
a
is connected. This implies
any a A is contained in some component of A.
Moreover, if we are given two components, say C
1
and C
2
, of A such that
C
1
C
2
is nonempty, take a C
1
C
2
. Then C
a
C
1
since a C
1
and C
1
is
connected. But C
a
is maximal by denition, so C
1
C
a
. Hence C
1
= C
a
. A
similar argument shows C
2
= C
a
, so C
1
= C
2
. Hence any two components are
either disjoint or equal.
3.7 Theorem. The components of an open set in C are open.
Proof. Let A be an open set of C. If C is a component of A and a C, then
there exists an open disk around a in A. This disk is connected and contains a,
so the disk must lie in C. Hence C is open.
The only property of C that we used is that C is locally connected. So
the theorem can be generalized where C is replaced with any locally connected
topological space X.
Another consequence of our work on connectedness is the following.
3.8 Theorem. If f is analytic on a connected set A and f

0, then f is
constant.
To say that f is analytic on a set A, then we mean f is analytic on some
open U C that contains A.
Proof. Since f is analytic on an open connected set containing A and f

0, we
can connect points using polygonal paths with sides parallel to the coordinate
axes and use the real integrals of u
x
, v
x
, u
y
and v
y
, which are all zero.
Material for Later 33
3.2 Arc and Curves
3.9 Denition. An arc in C is a continuous map : [0, 1] C.
3.10 Remark. It is important to note that some of the literature refers to an
arc as the image of [0, 1] in C under , but others refer to an arc as the map
itself.
3.11 Denition. A Jordan arc is an arc such that is injective.
3.12 Denition. A closed curve is an arc with (0) = (1).
3.13 Denition. An arc is a dierentiable arc if

exists and is continuous.


It is regular if also

,= 0.
The corresponding denitions of Jordan curves, dierentiable curves, and
regular curves are all obvious.
One can reparameterize an arc by taking any continuous bijection of [0, 1]
with itself and looking at . If we have a regular arc, then the reparameterized
arc is regular if and only if is dierentiable with continuous and nonzero
derivative.
3.14 Denition. A piecewise dierentiable arc is one where

exists and is
continuous except at a nite number of points where the left and right derivatives
exist and are not zero.
If is the arc denoted by (t) for t [0, 1], then is the arc traversed
backwards, namely = (1 t). Any arc of the form [z a[ = r is always
given the parameterization a + re
i
: [0, 2) C. That is, we traverse any
circle in the counterclockwise direction.
CHAPTER 4
Conformal Mappings
4.1 Linear Fractional Transformations
Let
1
and
2
be regular curves intersecting at some point z =
1
(t
1
) =
2
(t
2
).
Then the angle between
1
and
2
at z is dened to be the angle between their
tangent lines. By convention, it is measured from
1
to
2
. This is easy to
calculate. The tangent line to
1
(t
1
) at z is given by

1
(t
1
), and similarly for

2
(t
2
), so the angle is given by
= arg
_

2
(t
2
)
[

2
(t
2
)[
_

1
(t
1
)
[

1
(t
1
)[
_
1
_
.
4.1 Denition. A map f : C C is conformal if it preserves angles between
regular curves.
Suppose now that we have an analytic function f : C C and look at the
angle between (regular) curves f
1
and f
2
at f(z). Using the chain rule,
we have
angle = arg
_
(f


2
)(t
2
)

2
(t
2
)
[(f


2
)(t
2
)

2
(t
2
)[
[(f


1
)(t
1
)

1
(t
1
)[
(f


1
)(t
1
)
1
(t
1
)
_
= arg
_

2
(t
2
)
[

2
(t
2
)[
[

1
(t
1
)[

1
(t
1
)
_
,
35
36 Linear Fractional Transformations
provided f

(z) ,= 0. This yields the following.


4.2 Theorem. If f is an analytic, injective function and f

,= 0, then f is
conformal.
A converse of this theorem is valid provided the rst order real and imaginary
parts of the f have continuous derivatives in their real and imaginary directions.
That is, if f = u(x, y) + iv(x, y) is conformal and such that u
x
, v
x
, u
y
and v
y
are all continuous, then the Cauchy-Riemann equations are satised, so f is
analytic.
The most basic examples of conformal mappings are the following
f(z) = az for some a C. One often thinks of this as the composition
of f(z) = kz, for some k > 0, which is a dilation or homothety, and
f(z) = e
i
z, which is rotation.
Translation by b C is a conformal map. That is, f(z) = z + b is
conformal.
The inversion map f(z) = 1/z is conformal. One should think of f on

C.
It is obvious that the map is conformal at all nonzero z. (The conformality
at z = 0 is satised almost by denition because we dened the behavior
of functions at in the Riemann sphere by using inversion.)
Compositions of these basic conformal maps are also conformal on all of

C.
Conjugation is not a conformal map.
4.3 Lemma. Compositions of the examples above have the form
f(z) =
az +b
cz +d
,
for some a, b, c, d C.
Since this map is unchanged by taking a a, b b, c c, and d d,
we may normalize the coecients. The usual normalization is adbc = 1. This
also eliminates the trivial cases, such as z d, etc. The condition ad bc ,= 0
actually eliminates all the trivial cases.
Proof. All of the basic maps (dilation, translation, inversion) are of this form,
and we may prove that this collection of maps forms a group, called the Mobius
Conformal Mappings 37
group. Let
f(z) =
az +b
cz +d
, g(z) =
a

z +b

z +d

.
Then
(f g)(z) =
a(
a

z+b

z+d
) +b
c(
a

z+b

z+d
) +d
=
(aa

+bc

)z + (ab

+bd

)
(ca

+c

d)z + (cb

+d

d)
.
Observe
(f g)(z) =
a

z +b

z +d
,
where
_
a

_
=
_
a b
c d
__
a

_
.
Since ad bc = a

= 1 and the case a

= d

= 1 and b

= c

= 0, i.e.,
the identity matrix, corresponds to f(z) = z. We conclude that our maps form
a group and, in fact, that
az +b
cz +d

_
a b
c d
_
is a group isomorphic with SL
2
(C).
So we have shown more than what the lemma told us. We have shown
additionally that we have a group, and not just a monoid (a group without
inversion). We note that SL
2
(C) is the set of analytic isomorphisms of

C. It
can be shown that the analytic isomorphisms of R are isomorphic to SL
2
(R),
and these, in turn, the analytic automorphisms of the upper half-plane H are
isomorphic to SL
2
(R).
Additionally, we can write the maps using matrices directly. Think of z C
as a point (z
1
, z
2
) with homogenous coordinates so z
1
/z
2
= z. (Notice the
correlation to the complex projective line P
1
(C).) Then
f(z) =
_
a b
c d
__
z
1
z
2
_
.
Thus we have a group of maps generated by our basic examples (dilations,
translations, and inversion), which is a subgroup of SL
2
(C) = M, where M is
the Mobius group.
38 Features of M obius Transformations
4.4 Theorem. The Mobius group is generated by the translations, dilations,
and inversion. In fact, any f M may be written as a composition
f(z) =
1

2
,
where
1
and
2
are translations, is inversion, and is dilation.
Proof. We have
az +b
cz +d
=
a
c
(cz +d)
da
c
+b
c(z +
d
c
)
=
a
c

ad bc
c
2
(z +
d
c
)
=
a
c

1
c
2
(z +
d
c
)
.
So translate by z z +
d
c
, invert by z
1
z
, multiply by
1
c
2
, then translate by
z z +
a
c
. These maps our the desired
2
, , , and
1
, respectively.
These maps correspond to movements of the Riemann sphere. Indeed, trans-
lation of the north pole at xed height (i.e., parallel to the C-plane) gives trans-
lations, raising the sphere up gives dilations, and rotations of the sphere around
the axis perpendicular to C gives rotations. Finally, rotation through an angle
around an axis parallel to C gives 1/z.
4.5 Example. We show that given any element z = x + iy in the upper half-
plane H, there is a matrix (
a b
c d
) in SL
2
(R) such that f(z) =
az+b
cz+d
maps i to z.
Indeed, following [SL99, p. 124], it suces to show such (
a b
c d
) exists in GL
+
2
(R),
as we can nd R such that (
a b
c d
) SL
2
(R). Note that a matrix of the
form (
1 b
0 1
) corresponds to a translation by b. So we may translate z to a point
on the imaginary axis, say ri with r > 0. Then the matrix (
1 0
0 r
) maps ri to
i and dene the desired linear fractional transformation by (
1 0
0 r
)(
1 b
0 1
) = (
1 b
0 r
).
Moreover, given any z
1
and z
2
in C, there exists a matrix (
a b
c d
) in SL
2
(R) such
that f(z) =
az+b
cz+d
takes z
1
to z
2
. Namely, use the matrix taking z
1
to i, then
follow that by the matrix taking i to z
2
.
4.2 Features of Mobius Transformations
4.6 Theorem. Linear fractional transformations map the set of circles and
lines to the set of circles and lines.
Proof. The basic maps already have this property. However, one must prove
this for the inversion map, which is left as an exercise.
Conformal Mappings 39
4.7 Theorem. Given any three distinct points z
1
, z
2
, and z
3
, there exists
a linear fractional transformation T such that T(z
1
) = 1, T(z
2
) = 0, and
T(z
3
) = .
Proof. Send
z
z z
2
z z
3
z
1
z
3
z
1
z
2
.
This can then be written in the usual form (
a b
c d
).
4.8 Theorem. Any linear fractional transformation is determined by the im-
ages of three distinct points.
Proof. Without loss of generality (using the previous theorem), we know T(z
1
) =
1, T(z
2
) = 0, and T(z
3
) = , and we know a linear fractional transformation
that has this property, call it g. Then g T
1
xes 0, 1, and , and is a linear
fractional transformation. So
(g T
1
)(z) =
az +b
cz +d
.
Then (g T
1
)(0) = 0 implies b = 0, (g T
1
)() = implies c = 0, and
(g T
1
)(1) = 1 implies a/d = 1. Hence g T
1
= id.
4.9 Denition. Let z
1
, z
2
, z
3
C. Then
(z, z
1
, z
2
, z
3
) =
z z
2
z z
3
z
1
z
3
z
1
z
2
= f(z),
where f takes z
1
, z
2
, z
3
to 1, 0, . This is called the cross ratio of (z, z
1
, z
2
, z
3
).
4.10 Theorem. The cross ratio is invariant under linear fractional transfor-
mations.
Proof. Let g be an linear fractional transformation. Look at (g(z)g(z
1
), g(z
2
), g(z
3
)).
Since f g
1
maps g(z
1
) 1, g(z
2
) 0 and g(z
3
) , we have
(g(z), g(z
1
), g(z
2
), g(z
3
)) = (f g
1
)(g(z)) = f(z),
as required.
4.11 Theorem. A cross ratio is real if and only if the points lie on a circle or
line.
40 Reflection in a Circle
Proof. Take four points and assume their cross ratio is real. A linear fraction
transformation maps the last three points to 1, 0, and . The cross ratio is still
real and is not equal to the rst point, so the points are all in R. An inverse
linear fractional transformation takes R to a line or circle.
4.12 Example. What if we want to send three points to points dierent from
1, 0, and ? In particular, what is the linear fractional transformation sending
0, 1, and 2 to i, 1, and 1? It is [SL99, p. 141]
f(z) =
(1 + 3i)z 4i
(i + 3)z 4
.
4.3 Reection in a Circle
We know how to reect in R, z z, so we can dene reection in a circle.
Indeed, given a circle, a linear fractional transformation takes the circle to R,
then apply z z, and apply another linear fractional transformation to take R
back to the circle.
Consider the circle of radius R centered at a C; i.e., z C : [z a[ = R.
We can compute the reection as follows. Suppose z
1
, z
2
, and z
3
are points
on the circle. Then (z, z
1
, z
2
, z
3
) is the image of z under a linear fractional
transformation taking the circle to R. So (z, z
1
, z
2
, z
3
) is a reection in R. If we
could nd w such that (w, z
1
, z
2
, z
3
) = (z, z
1
, z
2
, z
3
), then w is the reection of
z in the circle. Then
(z, z
1
, z
2
, z
3
) = (z a, z
1
a, z
2
a, z
3
a)
=
_
z a,
R
2
z
1
a
,
R
2
z
2
a
,
R
2
z
3
a
_
=
_
R
2
z a
, z
1
a, z
2
a, z
3
a
_
=
_
R
2
z a
+a, z
1
, z
2
, z
3
_
.
So the reection is
z
R
2
z a
+a.
Conformal Mappings 41
4.4 Dynamics
We can identify two special types from our basic maps, namely z z + b and
z az, for a, b C. The reason these are special is that under each
map, and z az also xes zero. Additionally, these statements are if and only
if. It turns out every linear fractional transformation looks like these in an
appropriate coordinate system. Indeed, a xed point of
f(z) =
az +b
cz +d
occurs at
z =
az +b
cz +d
,
that is when cz
2
+ (d a)z b = 0. This gives two solutions or one solution
with multiplicity two.
In the two solution case, let and be the xed points of f. If we take
a linear fractional transformation g which sends 0 and , then
g
1
f g is a linear fractional transformation which xes 0 and . Therefore,
it is a dilation and rotation; i.e., z z for some C. If we look at the
polar grid, it behaves nicely under z z. Notice grid circles go to grid circles,
and rays go to rays. In the (very) special case where is real and positive, rays
are preserved and grid circles are expanded or contracted. If [[ = 1, then rays
are rotated and circles are preserved. So, z z is a composition of the latter
two cases. If we push this grid forward via g, then g(grid rays) gives a family
Figure 4.1: The polar grid in C.
of circles through the xed points and , and g(grid circles) gives a family of
42 Dynamics
circles separating and which is orthogonal to g(rays).
Figure 4.2: The push-forward of the polar grid under g. This type of set-up is
called hyperbolic dynamics. If > 0 then is attracting and is repelling, and if
< 1 then is attracting and is repelling. The case where || = 1 is called the
elliptic case. In either case, we call the diagram the Steiner circle.
If we have only one xed point (with multiplicity 2), then we can do (approxi-
mately) the same thing, mapping the xed point to and using the appropriate
dilation to get g
1
f g is z z + for some C. There is a natural grid,
shown in Figure 4.3. The push-forward of the grid under g gives parabolic
Figure 4.3: The grid with lines parallel and perpendicular to the vector .
behavior.
Conformal Mappings 43
Figure 4.4: When z z + , we get a parabolic scenario.
4.5 Conformal Mappings of a Simply Connected
Domain
4.13 Denition. A domain is a connected open set.
4.14 Denition. A domain in C is simply connected if its complement (in

C)
has at most one component.
There are two pictures to keep in mind. Something homeomorphic to a disk
is simply connected, but something with a hole in it is not; i.e., an annulus.
We will see much better denitions of simple connectedness when we talk about
integration theory.
4.15 Riemann Mapping Theorem. If

C is a simply connected domain
and

C contains at least two points, then there exists an analytic bijection
f : D. If we specify a point z , we can require that f(z) = 0. If we
specify that f(z) = 0 and f

(z) > 0, then f is unique.


We will prove this much later in the course, but the result is useful to see
here so as to make clear the motivation for other topics, such as linear fractional
transformations.
This begs the question: How does one obtain such maps? The answer is:
I dont know! The proof is purely existential and no constructive proof exists.
If you nd a constructive proof, Professor Rogers will get you a postdoc at a
top university.
In certain simple cases, however, it is possible to get these maps.
44 Conformal Mappings of a Simply Connected Domain
4.16 Example. Consider
z
z i
z +i
.
This sends the upper half-plane H to the unit disk D such that i 0.
Figure 4.5: We can map the upper half-plane to the unit disk. So, one can map any
half-plane to the unit disk by using rotations of the Riemann sphere.
4.17 Example. Consider the sector in C with angle

. Then the map z z

sends this region to the upper half-plane. Writing z = re


i
with 0 < <

and
0 < r < , we have z

= r

e
i
.
4.18 Example. Let N = z C : Re(z) < 0, Im(z) = 0. Then we can map
CN H via z

z. This map, of course, is just a special case of a power


map.
From these examples, we can extract the following idea. The power map
opens or closes a positive angle at 0 and . As a consequence, if we are given a
region enclosed by two circles intersecting with angle

(see Figure 4.6), we can


obtain a sector by a linear fractional transformation, then rotate to a sector,
then apply z z

to get the upper half-plane.


4.19 Example. Let = z C : 0 < Im(z) < . Then z e
z
takes to
H. If we consider

= z C : 0 < Im(z) <


for some > 1, then z e


z
sends

to a sector with angle


. Hence z e
z
sends

H. So the
exponential map opens up a zero angle at to positive size. That is, on the
Riemann sphere, we have a strip that maps around the sphere and at the south
pole we have some thickness. However, the strip narrows and becomes a cusp
at .
Conformal Mappings 45
Figure 4.6: A general conformal map.
Another exponential conformal map is the map in which a strip is taken to
C z C : Re(z) < 0, Im(z) = 0. Let = z C : i < Im(z) < i.
Then e
z
is our desired map.
4.20 Example. Consider the region given in Figure 4.7. Then a linear
fractional transformation sends 1 to 0 and 1 to 0. So z
z+1
z1
. This map
takes us to a quadrant. Then squaring it gives a half-plane and rotating gives
us the upper half-plane. Hence
z
_
z + 1
z 1
_
2
: H.
If we move 0 to 1 and to 1 so that 1 and 1 are preserved by the map,
then this is the same as composing with
z
az +b
cz +d
.
So send
z
z + 1
z + 1
.
46 Conformal Mappings of a Simply Connected Domain
Figure 4.7: The map z (
z+1
z1
)
2
takes the H {z D : Im(z) > 0} to H.
So the map is now
z
(z1)
2
(z+1)
2
+ 1
1 (
z1
z+1
)
2
=
(z 1)
2
+ (z + 1)
2
(z + 1)
2
(z 1)
2
=
2z
2
+ 2
4z
=
1
2
_
z +
1
z
_
.
So the map z
1
2
(z +
1
z
) takes the region in Figure 4.7 to H, where both the
points 1 and 1 are preserved.
Observe that this map has reective symmetry in the real axis; i.e., if f(z) =
1
2
(z +
1
z
), then
f(z) =
1
2
_
z +
1
z
_
=
1
2
_
z +
1
z
_
= f(z).
Consequently, the function f maps C D C ([1, 1] R). But what
happens to D? Recall, the map D CD is dened by z
1
z
. The function
f above is unchanged by z
1
z
, so the same map f by z
1
2
(z +
1
z
) sends
D C ([1, 1] R).
4.21 Example. Consider the half-strip z C : 0 < Im(z) < , Re(z) > 0.
Then z e
z
takes the half-strip to the region in Figure 4.7. Then applying
z
1
2
(z +
1
z
) takes the region into C ([1, 1] R). But this composition
[z
1
2
(z +
1
z
)] [z e
z
] = [z cosh z].
Weve seen some useful properties of the map z
1
2
(z +
1
z
). What else can
we do with it? In particular, are there any curves that have a nice form under
this map? Indeed,
x +iy
1
2
_
x +iy +
x iy
x
2
+y
2
_
=
x
2
_
1 +
1
x
2
+y
2
_
+
iy
2
_
1
1
x
2
+y
2
_
.
Conformal Mappings 47
We see immediately that the circle x
2
+ y
2
= R
2
is mapped to an ellipse. In
particular, we have Figure 4.8.
Figure 4.8: The image of concentric circles about the origin under the mapping
z
1
2
(z +
1
z
).
Can we do some of the same things with Mobius maps? Yes! Consider the
(global) conformal map z
zi
z+i
which takes H D. Any line through 0
and must go to a curve through 1 and 1. As a special case, the imaginary
axis goes to [1, 1], and the rst quadrant goes to the lower hemisphere of D.
Also, we can map the sector with angle

and bounded by [z[ = 1 to the upper


hemisphere of D via z z

, then invert to get the lower hemisphere of D, then


to the third quadrant, then to all of H.
The following are some more pictorial examples of conformal mappings.
Figure 4.9: Using a linear fractional transformation, we send A 0, B 1 and C
goes to another point on S
1
. The circle through A and B maps to R and the circle
through B and C goes to S
1
.
4.22 Example. Let z
0
be an arbitrary complex number. Then there is a linear
fractional transformation such that z . Indeed, following [SL99, p. 137], let
48 Conformal Mappings of a Simply Connected Domain
Figure 4.10: The image of the right half-plane, including the line Re(z) = 1, onto
the entire complex plane C.
A = (
0 1
1 z0
) GL
2
(C). Then the associated linear fractional transformation
f(z) =
1
z z
0
sends z
0
.
We also include some mappings from additional sources. There is a complex
function viewer available which displays functions as in Figure 4.11. Addition-
ally, we include a table of conformal mappings from [Fis90, pp. 389-394]. In
these tables, let z = x+iy and w = u+iv. The lowercase letters and the cross-
hatched regions in the z-plane map to capital letters and the shaded regions in
the w-plane, respectively.
Conformal Mappings 49
Figure 4.11: We can see what the map z z
2
does to any rectangle about the
origin in C. Picture courtesy of Bombelli [CF].
50 Conformal Mappings of a Simply Connected Domain
Figure 4.12: The mappings z e
z
, z z

for > 0, and z sin(


z
2
)
from [Fis90, p. 389].
Conformal Mappings 51
Figure 4.13: The mappings z sin z, z log z, and z z
2
from [Fis90, p. 390].
52 Conformal Mappings of a Simply Connected Domain
Figure 4.14: The mappings z
1
z
, z i(
1+z
1z
), and z i
z
2
+2iz+1
z
2
2iz+1
from [Fis90, p. 391].
Conformal Mappings 53
Figure 4.15: The mappings z
az
1az
, z cosh(
z
h
), z

2
(z +
1
z
), z (z +
1
z
),
and z z + e
z
from [Fis90, p. 392].
54 Conformal Mappings of a Simply Connected Domain
Figure 4.16: Some elaborate conformal mappings from [Fis90, p. 393].
Conformal Mappings 55
Figure 4.17: The mappings z 2

z + 1 + log(

z+11

z+1+1
) and z
z
1z
, where
=
1+ab

(1a
2
)(1b
2
)
a+b
and R =
1ab

(1a
2
)(1b
2
)
ba
, from [Fis90, p. 394].
CHAPTER 5
Integration Theory
5.1 Line Integrals
Let f = u +iv be a continuous map [0, 1] C. Then
_
1
0
f(t) dt =
_
1
0
u(t) dt +i
_
1
0
v(t) dt.
If we have a dierentiable curve , parametrized by z : [0, 1] C, z

is contin-
uous, and f : C C is continuous, then we dene
_

f(z) dz =
_
1
0
f(z

(t))z

(t) dt.
If z = x +iy and f = u +iv, then
_

f(z) dz =
_
1
0
(u +iv)(x

+iy

) dt =
_
1
0
(ux

vy

) dt +i
_
1
0
(vx

+uy

) dt,
or as in (possibly more familiar) real variable notation
_

f(z) dz =
_
1
0
(udx v dy) +i
_
1
0
(v dx +udy).
57
58 Line Integrals
We need to verify that the integral of f along does not depend on the choice
of parameterization; i.e., that the integral is well-dened. Suppose we have the
reparameterization : [0, 1] [0, 1] with

continuous and

> 0. Then
_
1
0
f((z )(t))(z )

(t) dt =
_
1
0
f((z )(t))(z

)(t)

(t) dt
=
_
1
0
f(z(s))z

(s) ds
=
_

f(z) dz,
where s = (t).
Now that we have a well-dened line integral, we can look at some of its
properties. For any , C and functions f and g, we have
_

(f +g) dz =
_

f dz +
_

g dz.
The proof of which is done by converting each integral into its real and imaginary
parts and using the denition, so the proof is left to the reader. This shows the
integral is linear. The integral is also monotone. Indeed, if g : [0, 1] C is
continuous, then

_
1
0
g(t) dt

_
1
0
[g(t)[ dt.
Proof. Let e
i
be such that
e
i
_
1
0
g(t) dt =

_
1
0
g(t) dt

.
Then
_
1
0
e
i
g(t) dt =

_
1
0
g(t) dt

by linearity of the integral. But


_
1
0
e
i
g(t) dt = Re
__
1
0
e
i
g(t) dt
_
=
_
1
0
Re(e
i
g(t)) dt.
Note Re(e
i
g(t)) [e
i
g(t)[, so using the properties of the real integral, we have
_
1
0
Re(e
i
g(t)) dt
_
1
0
[e
i
g(t)[ dt =
_
1
0
[g(t)[ dt
Integration Theory 59
as required.
Further, we can show

f(z) dz

[f(z)[ [dz[ =
_

[f(z(t))[ [z

(t)[ dt,
where [dz[ is the arc length measure and z is some parameterization of .
Proof. We have
_

f(z) dz =
_
1
0
f(z(t))z

(t) dt,
so taking absolute value and using the previous property we have the desired
result.
Lastly,
_

f(z) dz =
_

f(z) dz,
where is with the reverse orientation. That is, if z(t) parametrizes , then
z(1 t) parametrizes and we have
_

f(z) dz =
_
1
0
f(z(1t))z

(1t) dt =
_
1
0
f(z(s))(z

(s)) ds =
_

f(z) dz.
Also, we note there is an obvious extension to any piecewise dierentiable
arc .
We look now for some sort of fundamental theorem for line integrals. Re-
call from calculus Greens theorem, which says that a path-independent integral,
namely
_

p dx +q dy
from (a, b) to (c, d) in R
2
exists if and only if there exists W such that dW =
p dx +q dy. Said dierently, the integral
_

V dr
is path independent if and only if V = W for some W.
60 Line Integrals
5.1 Greens Theorem (Weak version). Let p and q be continuous. The integral
_

p dx +q dy
depends only on the endpoints of if and only if there exists W such that
p =
W
x
, q =
W
y
.
If so, then
_

p dx +q dy = W(z(1)) W(z(0)),
where z : [0, 1] C is a parameterization of .
Proof. If there exists such W, then
_

p dx +q dy =
_

W
x
dx +
W
y
dy
=
_
1
0
_
W
x
dx
dt
+
W
y
dy
dt
_
dt
=
_
dW
= W(z(1)) W(z(0)).
So the integral depends only on the endpoints.
Conversely, suppose
_

p dx +q dy
depends only on the endpoints of . Fix a basepoint z
0
. Dene W(z) by
W(z) =
_

p dx +q dy
for with starting point z
0
and endpoint z. Then
W(z +h) W(z) =
_
1
p dx +q dy
_
2
p dx +q dy.
Choose
1
and
2
such that
2
is
1
followed by a line segment from z to z +h.
Integration Theory 61
Then
W(z +h) W(z)
h
=
1
h
_
line segment
p dx +q dy
=
1
h
_
1
0
(p Re(h) +q Im(h)) dt.
We only want W
x
, which corresponds to h real, and W
y
, which corresponds to
h imaginary, so
W
x
=
W
x
= lim
h0
1
h
_
1
0
p(ht)hdt = p.
Similarly, W
y
= q.
What does this have to do with complex functions? Well since
_

f(z) dz =
_

(u +iv) dx +i(u +iv) dy,


our reasoning says the following.
5.2 Theorem. Let be a curve in a region . The integral
_

f(z) dz
is path-independent if and only if there exists an analytic function F with F
x
=
u + iv = f and F
y
= i(u + iv) = if; i.e., there exists a function F analytic in
such that F

= f.
Proof. Directly apply the previous theorem to get the existence of F with con-
tinuous partial derivatives F
x
= f and F
y
= if. To get F analytic, it suces
that the Cauchy-Riemann equations hold. This is clear because

x
Re(F) = Re(f) = u and

y
Im(F) = Im(if) = u
and

x
Im(F) = Im(f) = v and

x
Re(F) = Re(if) = v.
So the Cauchy-Riemann equations hold.
5.3 Corollary. The integral of a polynomial in C[z] along a path is path-
independent. In particular, the integal of a polynomial in C[z] along a closed
62 Line Integrals
curve is zero.
5.4 Corollary. The function f(z) = 1/z has no antiderivative on an open set
containing zero.
Proof. If it did, then
_

1
z
dz = 0
for any closed curve . So in particular, the integral is zero for a small circle
around zero. But we can parameterize such a circle by z(t) = re
2it
for t [0, 1]
and r small. Then
_

1
z
dz =
_
1
0
1
re
2it
2ire
2it
dt =
_
1
0
2i dt = 2i ,= 0.
Hence no antiderivative exists.
Now for some motivation. Observe that by Greens theorem, if W is C
2
,
then
p
y
=
q
x
=

2
W
xy
.
This gives Greens theorem for a rectangle.
5.5 Greens Theorem (Rectangle Version). Let R be a rectangle in R
2
and
p: R R
2
and q : R R
2
be C
1
functions. If p
y
= q
x
in R, then
_
R
p dx +q dy = 0.
Proof. Lets see a small picture of why this is true. Suppose p
y
= q
x
on a very
small rectangle. Since the rectangle is very small, we have
p(x, y) = p(0) +p
x
(0)x +p
y
(0)y +o()
and
q(x, y) = q(0) +q
x
(0)x +q
y
(0)y +o().
Observe that
_

c dx =
_

c dy = 0,
for any constant c. Also,
_

o() dx +o() dy = o(
2
).
Integration Theory 63
Figure 5.1: This shows geometrically why Greens theorem is true on a rectangle.
Then
_
/2
/2
xdx =
x
2
2
= 0
and
_
/2
/2
y dy =
_
/2
/2

2
dx +
_
/2
/2

2
=
2
.
Similarly
_
y dy = 0 and
_
xdy =
2
.
So
_

p dx +q dy = p
y
(0)(
2
) +q
x
(0)
2
+o(
2
) =
__
R
(q
x
p
y
) dA+o(
2
),
where R is the rectangle.
So what we have done is considered a small rectangle

R with side-lengths .
Using the Taylor expansions of p and q we found
_

e
R
p dx +q dy = o(
2
).
Now for the rectangle R, subdivide it into small rectangles R
j
. Suppose we have
N
2
rectangles. Then

_
R
p dx +q dy

N
2

j=1
_
Rj
p dx +q dy

N
2
o
_
c
2
N
2
_
.
Therefore, since o((
c
N
)
2
) 0 as N , we have the desired result.
64 Line Integrals
The complex variable version of the fundamental theorem of calculus is
Cauchys theorem for a rectangle (or the Cauchy-Goursat theorem).
5.6 Cauchys Theorem (For a Rectangle). Let R be a rectangle in C; that is,
let R = z = x +iy : a x b, c y d. If f is analytic on R, then
_
R
f(z) dz = 0,
where R consists of line segments given by
z(t) :=
_

_
y = c, x = a +t(b a)
x = b, y = c +t(d c)
y = d, x = b +t(a b)
x = a, y = d +t(c d)
for 0 t 1.
Proof (Goursat). Subdivide the rectangle into four subrectangles. Since
_
R
f(z) dz =
4

j=1
_
R1,j
f(z) dz,
there exist subrectangles such that

_
R1,j
1
f(z) dz

1
4

_
R
f(z) dz

.
Inductively subdividing R
n,jn
we see that there exists a sequence R
n,jn
of
rectangles of length scale 2
n
, so

_
Rn,jn
f(z) dz

1
4
n

_
R
f(z) dz

,
and R
n,jn
R
n1,jn1
for all n. In particular

n1
R
n,jn
= z
0
since it is
an intersection of nested compact sets with diameter approaching zero. Take
> 0. At z
0
, use the dierentiability of f to write

f(z) f(z
0
)
z z
0
f

(z
0
)

<
Integration Theory 65
for all [z z
0
[ suciently small. So in particular for large enough n,
[f(z) f(z
0
) f

(z
0
)(z z
0
)[ diam(R
n,jn
) c(2
n
).
for all z R
n,jn
. Thus

_
Rn,jn
_
f(z) f(z
0
) f

(z
0
)(z z
0
)
_
dz

8c
2
(2
n
)
2
.
But
_
Rn,jn
f(z) dz = 0
and
_
Rn,jn
f

(z
0
)(z z
0
) dz = 0
because both integrands have antiderivatives, so
4
n

_
Rn,jn
f(z) dz

8c
2
4
n
4
n
.
Thus

_
R
f(z) dz

8c
2

and > 0 is arbitrary.


In fact, we can say more.
5.7 Theorem. Suppose R C is a rectangle, f is analytic on R except at a
nite number of points
1
, . . . ,
n
, and lim
zj
(z
j
)f(z) = 0. Then
_
R
f(z) dz = 0.
Proof. Fix > 0. Subdivide the rectangle so that there is at most one
j
in
each subrectangle, where
j
is the center, and such that [z
j
[[f(z)[ on
the boundary of the rectangle containing
j
. Then

_
R
f(z) dz

k
_
R
k
f(z) dz

,
where k indexes all the subrectangles of R. But the subrectangles not containing
66 Line Integrals
a
j
are zero by the previous theorem, so

_
R
f(z) dz

j=1
_
Rj
f(z) dz

n
c diam(R
j
)
length(R
j
) Cn,
where R
j
is the rectangle containing
j
.
5.8 Cauchys Theorem (For a Disk). If f is analytic on an open disk D and
is a closed curve in D, then
_

f(z) dz = 0.
Proof. Dene F(z) as follows. Let z
0
be the center of the disk. For any other
z in the disk, take a path from z
0
to z in the disk, consisting of one vertical
and one horizontal line segment. Let
F(z) =
_

f(z) dz.
Our function F is well-dened. Indeed, if
1
and
2
are two distinct such paths
from z
0
to z, then
1
followed by
2
is the boundary of a rectangle which is
strictly contained in D. So
_
1
f(z) dz +
_
2
f(z) dz =
_
R
f(z) dz = 0
by the previous theorem. Now we look at F

(z) by computing the partial deriva-


tives F
x
and F
y
. The partial derivative F
x
can be computed by using the path
that has rst vertical then horizontal segment. Using that
_
f(z) dz =
_
f(z) dx +if(z) dy,
we nd F
x
= f. Similarly, using the other path, F
y
= if. So
Re(F)
x
= Re(f) = Im(if) =
Im(F)
y
and
Im(F)
x
= Im(f) = Re(if) =
Re(F)
y
.
Then the Cauchy-Riemann equations hold and F

= f. Since f is the derivative


Integration Theory 67
of the analytic function F, we have
_

f(z) dz = 0
for all closed curves .
5.9 Cauchys Theorem (General Version For a Disk). Suppose f is analytic
on an open disk D except at a nite collection of points
1
, . . . ,
n
at which
lim
zj
f(z)(z
j
) = 0.
If is a closed curve in D that does not contain any
j
, then
_

f(z) dz = 0.
We can look at more general curves using the same idea as in the proofs
above. For example, consider the curve given by Figure 5.2. If f is an
Figure 5.2: A more general curve to which the ideas of Cauchys theorem may be
applied.
analytic function in the upper half-plane H, then
_

f(z) dz = 0,
for =

+
R
+
+
+

, where

= [R, ] and
+
= [, R] for increasing
x-values,
R
: Re
i
for increasing in [0, ], and

: e
i
for decreasing in
68 Line Integrals
[0, ]. This curve can be used in the evaluation of certain improper integrals.
5.10 Example. Consider
_

1 cos x
x
2
dx,
which can viewed as an improper Riemann integral and thus approximated by
lim
R
lim
0
_

R
1 cos x
x
2
dx +
_
R

1 cos x
x
2
dx.
Observe that the integrand is Re(
1e
iz
z
2
). So we can write the integral
lim
R
0
Re
_
_

1 e
iz
z
2
dz +
_
+
1 e
iz
z
2
dz
_
.
So by Cauchys theorem we have
lim
R
0
Re
_

1 e
iz
z
2
dz
_

R
1 e
iz
z
2
dz
_
.
But

R
1 e
iz
z
2
dz

R
[1 e
iz
[
[z
2
[
[dz[
and
[e
iz
[ = [e
i(x+iy)
[ = e
y
1,
for y > 0 on
R
, and [z
2
[ = R
2
on
R
. Hence
_

R
[1 e
iz
[
[z
2
[
[dz[
_

R
2
R
2
[dz[
2R
R
2

2
R
0, R .
For

1 e
iz
z
2
dz =
_

1 e
iz
z
2
dz,
use Taylor expansion on
e
iz
= 1 +iz +
(iz)
2
2!
+ =
1 e
iz
z
2
=
i
z
+
1
2!
+ .
In fact,

1 e
iz
z
2
+
i
z

1
Integration Theory 69
for suciently small . So

_
1 e
iz
z
2
+
i
z
_
dz

[dz[ = 0, 0,
and
_

i
z
dz =
_

0
i
e
i
ie
i
d = .
Hence

1 cos x
x
2
dx

lim
R
0

1 e
iz
z
2
+
i
z
dz

R
1 e
iz
z
2
dz

lim
R
0
_
+
2
R
_
= 0,
so
_

1 cos x
x
2
dx = .
In practice, however, one leaves out the gory details. Write
_

1 cos x
x
2
dx = lim
R
0
_

R
by Cauchy. Then
lim
0
_

= and lim
R
_

R
= 0,
which yields our result.
5.11 Example. Suppose f is an analytic function on the closed disk [za[ R.
Consider the integral
_
|za|=R
f(z)
z a
dz.
The integrand is not analytic in the disk since a is a singularity. However, in
the annulus 0 < [z a[ R, f is analytic. But Cauchys theorem is not valid
in an annular region. However, if we add a tube, as in Figure 5.3, we can apply
Cauchys theorem.
Let =
R
+
+
+

denote the keyhole curve, where


R
: [z a[ = R
with a small piece of size removed and

: [z a[ = with small piece of size


removed, and

and
+
are the sides of the tube. (Here, is the width of
70 Line Integrals
Figure 5.3: This keyhole region allows us to apply Cauchys theorem in an annular
region.
the tube.) Then
_

R
+
_

+
_
+
+
_

=
_

= 0.
Further, as 0 we have
_
+
+
_

0.
So
_
|za|=R
f(z)
z a
dz =
_
|za|=
f(z)
z a
,
for all 0 R. For small ,
[f(z) f(a)[ sup
|za|
[f

(z)[,
so
_
|za|=
f(z) f(a)
z a
dz sup
|za|
[f

(z)[
_
|za|=
1
[z a[
[dz[
2 sup
|za|
[f

(z)[ 0
Integration Theory 71
as 0. So
lim
0
_
|za|=
f(z)
z a
dz = lim
0
_
|za|=
f(a)
z a
dz
= lim
0
f(a)
_
2
0
1
e
i
ie
i
d
= 2if(a).
This yields the following theorem.
5.12 Cauchys Integral Formula (Disk Version). For a function f analytic
in a disk [z a[ R, we have
f(a) =
1
2i
_
|za|=R
f(z)
z a
dz.
Heuristically, all we used was that the curve wrapped around a once. We
would expect that
1
2i
_

f(z)
z a
dz = f(a)
whenever wraps around a exactly once. In [Ahl79], things initially look a
little backwards. First, they give the following denition.
5.13 Denition. For a piecewise dierentiable closed curve and a C a
point not on , let
n(, a) =
_

1
z a
dz.
This is called the winding number of around a.
The reason this quantity is called the winding number is because it counts
the angle increase in around a. An important way to thing of it is
_

dz
z a
=
_

d(log(z a)) =
_

d(log [z a[) +i d(arg(z a)).


The rst term in the summand does not contribute anything since we are inte-
grating along a closed curve. However, the argument function is multi-valued,
so the integral is the total change in the angle.
5.14 Lemma. The quantity n(, a) is a multiple of 2i.
72 Line Integrals
Proof. Indeed, let z(t) be a parameterization of , where z(t) is piecewise dif-
ferentiable. Then
_

1
z a
dz =
_
1
0
z

(t) dt
z(t) a
.
Dene
h(s) :=
_
s
0
z

(t)
z(t) a
dt.
Then
h

(s) =
z

(s)
z(s) a
at all but a nite number of points. Then
d
ds
e
h(s)
= h

(s)e
h(s)
=
z

(s)
z(s) a
e
h(s)
,
so
d
ds
_
e
h(s)
(z(s) a)
_
= z

(s)e
h(s)
+z

(s)e
h(s)
= 0
at all but nitely many points. This implies e
h(s)
(z(s) a) is a constant, so
e
h(s)
(z(s) a) = z(0) a. Hence
e
h(1)
=
z(0) a
z(1) a
= 1
because z(0) = z(1). Hence h(1) = 2ij for some j Z. Hence n(, a) is an
integer multiple of 2i.
5.15 Theorem. If a is outside the disk [z z
0
[ R, meaning [a z
0
[ > R,
and is the circle [z z
0
[ = R with the usual orientation, then n(, a) = 0.
Proof. Apply Cauchys theorem to
f(z) =
1
z a
,
which is analytic in [z z
0
[ < R. We conclude that
_

f(z) dz = 0 = n(, a),


as required.
5.16 Theorem. The winding number is constant on each component of C.
Integration Theory 73
Most of the work of the proof is done in the following lemma.
5.17 Lemma. n(, z) is constant on any line segment that does not intersect
. (In fact, we only need vertical and horizontal line segments.)
Proof. We have
n(, z) =
_

d
z
.
If the endpoints of the interval are a and b, then
a
b
C(, 0) for all
(in fact, for all in the complement of the line segment from a to b). Thus
log(
a
b
) is single-valued and analytic on , which implies
n(, a) n(, b) =
_

_
1
a

1
b
_
d = log
_
a
b
_

endpoints
of
= 0,
because is closed. Hence n(, a) = n(, b). Thus the argument is valid when-
ever the segment from a to b does not intersect .
Proof of Theorem 5.16. Any pair of points in a component of C are con-
nected by a path in the component, so not intersecting , consisting of nitely
many vertical and horizontal line intervals (because components are open and
connected). So n(, z) is constant on each interval. Hence n(, z) is the same
at both endpoints, and therefore n(, z) is constant on the component.
5.18 Theorem. Let f be analytic in a disk D (except at nitely many
j
, with
(z
j
)f(z) 0 as z
j
), be a closed curve in the disk such that doesnt
contain any
j
, and a D not on . Then
_

f(z)
z a
dz = n(, a)f(a).
Proof. Apply Cauchys theorem (Theorem 5.8 on page 66) to
f(z) f(a)
z a
.
We conclude that
_

f(z) f(a)
z a
dz = 0.
Moving the f(a) term to the right-hand side gives the desired result.
74 Applications of the Cauchy Integral Formula
Note that this is an improved version of our Cauchy integral formula in a
disk (Theorem 5.12 on page 71).
5.2 Applications of the Cauchy Integral Formula
We have the representation formula
f(z) =
1
2i
_
|z|=R
f()
z
d.
We can try to dierentiate both sides (inside the integral). We should get
f

(z) =
1
2i
_
|z|=R
f()
( z)
2
d.
and, more generally,
f
(n)
(z) =
n!
2i
_
|z|=R
f()
( z)
n+1
d.
The integral is a continuous function of z!
5.19 Theorem. If f is analytic on a disk [z z
0
[ < R, then
f

(z) =
1
2i
_
|z0|=R
f()
( z)
2
d
for all [z z
0
[ < R.
Proof. Let be the circle [ z
0
[ = R. We have
f(w) f(z)
w z
=
1
w z
_
1
2i
_

f()
w
d
1
2i
_

f()
z
d
_
=
1
2i
_

f()
( w)( z)
d.
The question is the following: Does this integral converge to the desired result
Integration Theory 75
as w z? So

f(w) f(z)
w z

1
2i
_

f()
( z)
2
d

1
2i
_

f()
_
1
( w)( z)

1
( z)
2
_
d

1
2
sup

[f()[
_

1
[ z[
[w z[
[ z[[ w[
[d[.
If [z z
0
[ < R, then dist(z, ) = d > 0. Take [w z[ <
d
2
, so dist(w, )
d
2
.
Then we get
1
2
sup

[f()[
_

1
[ z[
[w z[
[ z[[ w[
[d[

1
2
sup

[f(z)[ d
3
2 [w z[ length() 0
as w z.
An inductive version of the same computation proves the following.
5.20 Theorem. Under the same hypotheses of the previous theorem,
f
(n)
(z) =
n!
2i
_
|z0|=R
f()
( z)
n+1
d.
5.21 Corollary. If f is analytic on a domain , then f is innitely dieren-
tiable.
We can also consider bounds on analytic f using the Cauchy integral formula.
5.22 Cauchys Estimates. If f is entire and M
r
= sup
|z|=r
[f(z)[, then
[f
(n)
(0)[
M
r
n!
r
n
.
Proof. We have
f
(n)
(z) =
n!
2i
_
||=r
f()
( z)
n+1
d
with z = 0, so we get
[f
(n)
(0)[
n!
2i
M
r
r
(n+1)
2r =
M
r
n!
r
n
.
76 Taylor Series
5.23 Liouvilles Theorem. If f is analytic and bounded on C, then f is
constant.
Proof. In this case, write
[f

(z)[
M
r
by applying previous reasoning to the circle [ z[ = r. Send r to get
f

(z) = 0 for all z, so f is constant.


5.24 Fundamental Theorem of Algebra. If P(z) is a nonconstant polyno-
mial, then there exists a C such that P(a) = 0.
Proof. Suppose P(z) is nonzero everywhere. Then
1
P(z)
is analytic on C. But
1
P(z)
is bounded (since it approaches is zero as z and it is continuous), so
1
P(z)
is constant. This implies P(z) is also constant, which is a contradiction.
5.25 Moreras Theorem. If f is continuous in a region and
_

f(z) dz = 0
for all closed curves in , then f is analytic.
Proof. Since
_

f(z) dz = 0 for all closed curves, the integral is path indepen-


dent. So we can dene an analytic antiderivative
F(z) :=
_
z
z0
f(z) dz,
and then F

= f is analytic.
5.3 Taylor Series
We can also use the Cauchy integral formula to understand what happens when
f is analytic, except at a point a C with
lim
za
(z a)f(z) = 0.
We know that in this case, if we take to be the boundary of a disk containing
a and on which f is analytic, then
f(z) =
1
2i
_

f()
z
d, z ,= a.
Integration Theory 77
In particular, this is a well-dened function even when z = a. Let
g(z) =
1
2i
_

f()
z
d,
for all z interior to . Then g is an analytic function inside because our
previous work says
g
(n)
(z) =
n!
2i
_

f()
( z)
n+1
d,
and f(z) = g(z) at z ,= a, for all z inside . We have shown the following
theorem.
5.26 Theorem. Let U C be an open set containing a. If f is analytic on
U a and
lim
za
(z a)f(z) = 0,
then there exists an analytic function g on U such that g = f on U a. We
call a U a removable singularity of f.
There is a nice way to get Taylor expansions from this. We do this by
applying the theorem to
f
1
(z) =
f(z) f(a)
z a
,
where f is analytic in a neighborhood of a. Observe f
1
(z) has a removable
singularity at z = a. Since the analytic extension of f
1
is in particular continuous
(since it is analytic) at z = a,
f
1
(a) = lim
za
f
1
(z) = f

(a).
This says
f(z) = f(a) + (z a)f
1
(z),
which is the 0th order Taylor expansion with error term. The error term is given
by
(z a)f(z) = (z a)
1
2i
_
|a|=r
f
1
()
z
d,
for suciently small r such that f is analytic on [ a[ r and for z with
78 Taylor Series
[z a[ < r. Then
(z a)
1
2i
_
|a|=r
f
1
()
z
d = (z a)
1
2i
_
|a|=r
f() f(a)
( a)( z)
d.
Now
(z a)
_
|a|=r
f(a)
( a)( z)
d = f(a)
_
|a|=r
_
1
z

1
a
_
d.
But this is precisely 0 since both z and a lie in the circle, and the integrals
denote n(, z) = 2i and n(, a) = 2i, respectively. So
f
1
(z) =
1
2i
_
|a|=r
f()
( a)( z)
d.
We get any order of the Taylor expansion using this idea by induction. So
f
1
(z)(z a) = f(z) f(a) and f
2
(z)(z a) = f
1
(z) f
1
(a), and so on down to
f
n
(z)(z a) = f
n1
(z) f
n1
(a). Notice f
n
(z) has a removable singularity at
z = a for all n (provided f
n1
is continuous, which we will get by induction),
and
f(z) = f(a) + (z a)f
1
(z)
= f(a) + (z a)(f
1
(a) + (z a)f
2
(z))
=
.
.
.
= f(a) + (z a)f
1
(a) + + (z a)
n1
f
n1
(a) + (z a)
n
f
n
(z).
We discover f
(n)
(a) = n!f
n
(a). This is the Taylor expansion with error:
f(z) =
n1

j=0
f
(j)
(a)
j!
(z a)
j
+ (z a)
n
f
n
(z),
with
f
n
(z) =
1
2i
_
|a|=r
f
n
(z)
z
d,
Integration Theory 79
for z inside the disk. Then
_
|a|=r
f
n
()
z
=
_
|a|=r
f()
n1

j=0
f
(j)
(a)
j!
( a)
j
( a)
n
( z)
d
and the terms
f
(j)
(a)
_
|a|=r
( a)
jn
z
d = 0
for all 0 j n 1, and so
f
n
(z) =
1
2i
_
|a|=r
f()
( a)
n
( a)
d
for [z a[ < r.
5.27 Taylors Theorem. If f is analytic at a, then for all n 1 we have
f(z) =
n1

j=0
f
(j)
(a)
j!
(z a)
j
+f
n
(z)(z a)
n
,
with
f
n
(z) =
1
2i
_
|a|=r
f()
( a)
n
( z)
d
for r small enough that f is analytic on [z a[ r and the points z inside this
disk.
5.28 Remark. We can estimate the error as follows. Let
M = sup
|a|=r
[f()[.
Then
[f
n
(z)[
Mr
r
n
(r [z a[)
.
So
[(z a)
n
f
n
(z)[
Mr
r [z a[
_
[z a[
r
_
n
0
as n .
5.29 Corollary. On compact sets inside a disk on which f is analytic, the
80 Zeros of Analytic Functions
Taylor series converges uniformly to f as n .
5.4 Zeros of Analytic Functions
Look at a function f analytic at z = a. We know
f(z) =
n1

j=0
f
(j)
(a)
j!
(z a)
j
+ (z a)
n
f
n
(z),
where
f
n
(z) =
_
|a|=r
f()
( a)
n
( z)
d, [z a[ < r.
One of two things happen.
Option 1. There is some nonzero f
(j)
(a). Take the expansion so the error
term is the rst nonzero one. So f(z) = (z a)
n
f
n
(z) and f
n
(a) ,= 0 because
n!f
n
(a) = f
(n)
(a). Thus f(z) has a zero at a which looks like the zero of a
polynomial (contrast this idea with
3

x on R with a zero at x = 0). Also, f


n
(a) ,=
0 implies f
n
(z) ,= 0 on a neighborhood of a. Hence there is a neighborhood of
a on which f(z) has no zeros except the one at a; that is, a is an isolated root
of f.
Option 2. All the f
(j)
(a) = 0. Then f 0 in [z a[ < r. Indeed, write
f(z) = (z a)
n
f
n
(z). Then
[f
n
(z)[ [z a[
n

1
2i
_
|a|=r
f()
( a)
n
( z)
d

_
[z a[
r
_
n
sup
|a|=r
[f()[
r [z a[
0 as n on [z a[ < r.
5.30 Theorem. If f is analytic on a region and f
(n)
(a) = 0 for all n at
some a , then f 0 on .
Proof. Let E = z : f
(n)
(z) = 0 for all n. Then E is nonempty because
a E, and by previous result if z E, then f 0 in a neighborhood of z. This
Integration Theory 81
neighborhood of z is in E, which implies E is open. But
E =

n
z : f
(n)
(z) = 0
is an intersection of relatively closed sets, and so is relatively closed. So E is a
nonempty closed and open subset of , which is connected, so E = .
We can now conclude the following.
5.31 Theorem. If f is an analytic function on a region and f is not iden-
tically zero, then the zeros of f are isolated and of nite order.
5.32 Corollary. If f is analytic on a region and z : f(z) = 0 has an
accumulation point in , then f 0.
5.33 Corollary. An analytic function that is not identically zero has only
nitely many zeros in any bounded set.
5.34 Corollary. Let C be a region. If f and g are analytic on and
z : f(z) = g(z) has an accumulation point in , then f = g.
We can also use these techniques to study isolated singularities.
5.35 Denition. A function f has an isolated singularity at z = a if f is
analytic on 0 < [z a[ < r for some r > 0.
We already know if
lim
za
(z a)f(z) = 0,
then z = a is a removable singularity. The remaining possibility is that f(z)
is unbounded in 0 < [z a[ < r. We know some functions like this, namely
(z a)
n
, or more generally, a polynomial in (z a)
n
, so something like
c
1
(z a)
1
+ +c
n
(z a)
n
. We say f has a pole at z = a if there exists a
polynomial P(z) such that f(z) P(
1
za
) has a removable singularity at z = a.
In this case, we can write
f(z) =
g(z)
(z a)
n
,
for g analytic in a neighborhood of a and g(a) ,= 0.
5.36 Theorem. If f has an isolated singularity at z = a, then f has a pole at
z = a if and only if
lim
za
f(z) = .
82 Zeros of Analytic Functions
Proof. The only if direction is clear since
f(z) =
g(z)
(z a)
n
,
where g(a) ,= 0 and g is analytic. To prove the if direction, notice f(z)
as z a implies we can take r > 0 suciently small that [f(z)[ > 1 on
0 < [z a[ r. But then
1
f(z)
is analytic on 0 < [z a[ r. It has a removable
singularity at z = a, and the analytic extension of
1
f(z)
to [z a[ r is zero at
z = a by continuity, so
1
f(z)
= (z a)
n
g(z)
for some analytic function g(z) with g(a) ,= 0. This implies
f(z) =
1
g(z)(z a)
n
=
h(z)
(z a)
n
,
where h(z) =
1
g(z)
is analytic and nonzero in a neighborhood of a. Hence z = a
is a pole.
The remaining type of isolated singularity has
lim
za
f(z)
does not exist and f(z) is unbounded on 0 < [za[ < r. It is called an essential
singularity. For example, e
1/z
on 0 < [z[ < 1.
5.37 Casorati-Weierstrass Theorem. If f has an essential singularity at
z = a, then for any r > 0, f gets arbitrarily close to every point in C on
0 < [z a[ < r.
Proof. If not, there exists r > 0 and C such that [f(z) [ > 0 on
0 < [z a[ < r. But then
lim
za
f(z)
z a
= .
This implies (z a)
1
(f(z) ) has a pole at z = a. So (z a)
1
(f(z) ) =
(z a)
n
g(z) for some n and g(z) analytic and nonzero at z = a. Then
lim
za
(z a)
n
(f(z) ) = 0.
Integration Theory 83
Then
lim
za
(z a)
n
= 0,
so
lim
za
(z a)
n
f(z) = 0.
This says (z a)
n
f(z) has a removable singularity at z = a and in fact a zero
there. So (z a)
n
f(z) = (z a)
m
h(z) where m n and h is analytic at a.
Thus f(z) = (z a)
mn
h(z) has either a pole or a removable singularity at a.
So the limit
lim
za
f(z)
exists in

C. This is a contradiction with z = a being an essential singularity.
5.5 Counting Zeros of Analytic Functions
5.38 Theorem. Suppose f is analytic in an open disk D, and let z
j
be the
set of zeros of f, where the multiplicity of z
j
is m
j
. Let be a closed curve in
D. Then
_

(z)
f(z)
dz =

j
m
j
n(, z
j
).
Proof. Notice is a compact set in C, so it is contained in some D

, where D

is disk compactly contained in D. Then f has only a nite number of zeros in


D

, namely z
1
, . . . , z
k
, so we can write
f(z) =
_
k

j=1
(z z
j
)
mj
_
g(z)
and
f

(z)
f(z)
=
_
k

j=1
a
j
z z
j
_
+
g

(z)
g(z)
.
Integrating the latter of , we have
_

(z)
f(z)
dz =
_

_
k

j=1
m
j
z z
j
_
dz +
_

(z)
g(z)
dz =
k

j=1
m
j
n(, z
j
),
because g(z) ,= 0 on D

, so g

(z)/g(z) is analytic on D

and Cauchys theorem


84 Counting Zeros of Analytic Functions
(coming soon)
Figure 5.4: For a closed curve and an analytic function f with zeros z1, z2, z3, the
image f() wraps the origin once for each zj in .
(Theorem 5.9 on page 67) says
_

(z)
g(z)
dz = 0.
This is what we wanted to show.
This theorem makes rigorous the picture in Figure 5.4. Indeed, let w = f(z).
Then
k

j=1
m
j
n(, z
j
) =
_

(z)
f(z)
dz =
_
f()
dw
w
= n(f(), 0),
where the rst equality follows from the theorem, the second from the change
of variables rule for integrals, and the third by an observation from the picture.
So the theorem shows that the number of times f() winds around zero is equal
to the number of times winds around each z
j
(counting multiplicity).
5.39 Example. We can count the number of times f(z) = a by considering
f(z) a. If z
j
(a) is the set of zeros of f(z) a with multiplicity m
j
(a), then

j
m
j
(a)n(, z
j
(a)) =
_

(z)
f(z) a
dz =
_
f()
dw
w a
= n(f(), a).
5.40 Theorem. Suppose f is analytic at some point z
0
and f(z
0
) = w
0
. Let n
be the order of the zero of f(z) w
0
at z
0
; that is, f(z) w
0
= (z z
0
)
n
g(z),
for some analytic g with g(z
0
) ,= 0. If > 0 is suciently small that g(z) ,= 0
in [z z
0
[ , then there exists > 0 such that if [w w
0
[ < then f(z) = w
at n points (counting multiplicity) in [z z
0
[ .
What the heck does this theorem really say? Well, we have the picture in
Figure 5.5.
(coming soon)
Figure 5.5: The image of the circle |z z0| = is a curve that does not hit w0.
Proof. Let be the curve [zz
0
[ = and n(f(), w
0
) = n. In fact, n(f(), w) =
n for all w is a component of Cf() that contains w
0
. If we take > 0 small
Integration Theory 85
enough that [ww
0
[ < is in this component, then n(f(), w) = n, and letting
z
j
(w) = z : f(z) = w gives
n = n(f(), w) =

j
m
j
(w)n(, z
j
(w)).
That is, f(z) = w n times (counting multiplicity) in [z z
0
[ < because
n(, z) = 1 for [z z
0
[ < and is n(, z) = 0 if [z z
0
[ > .
5.41 Corollary. If f is analytic, then f is an open map.
Proof. To see that f takes open sets to open sets, it suces to show that the
image of any suciently small disk [z z
0
[ < contains a disk. We just saw
f(z) : [z z
0
[ < w : [w w
0
[ < .
5.42 Corollary. If f is analytic at z
0
and f

(z
0
) ,= 0, then there exists > 0
such that f is a homeomorphism on [z z
0
[ < .
Proof. In this case, we have n = 1, so f is injective on [z z
0
[ < .
5.43 Maximum Principle. If f is a nonconstant analytic function on a region
, then [f[ does not attain its maximum on .
Proof. If z
0
, then there exists > 0 and > 0 so that D(z
0
, ) and
f(D(z
0
, )) D(f(z), ). So there exists w with
[f(w)[ = [f(z)[ +

2
> [f(z)[,
as required.
5.44 Corollary. If f is analytic on a closed set B then the maximum of [f[ is
achieved on the boundary B. If it is achieved on an interior point of B, then
f is constant.
5.45 Schwarz Lemma. Let D be the open unit disk in C. Suppose f is analytic
on D, [f(z)[ 1, and f(0) = 0. Then [f(z)[ [z[ and [f

(0)[ 1. Moreover,
if there exists z D with [f(z)[ = [z[ or if [f

(0)[ = 1, then f(z) = cz for some


[c[ = 1; that is, f is a rotation.
Proof. Notice f(z)/z has a removable singularity at 0, so there exist a function
f
1
analytic on D such that f
1
(z) = f(z)/z on D 0 and f
1
(z) = f

(0) at
z = 0. Then
[f
1
(z)[
[f(z)[
[z[
.
86 General Form of Cauchys Theorem
On [z[ = 1 , we have
[f
1
(z)[
1
1
.
The maximum principle says that
[f
1
(z)[
1
1
in [z[ 1 . Taking 0 we get [f
1
(z)[ 1 for all z D. So for z ,= 0,
[f(z)[
[z[
1 =[f(z)[ [z[
and for z = 0, [f

(0)[ 1. Moreover, if there exists z such that [f(z)[ = [z[ or if


[f

(0)[ = 1, then equality holds in the maximum principle, so f(z)/z = c with


[c[ = 1.
5.6 General Form of Cauchys Theorem
Recall that we know if f is analytic on a rectangle R, then
_
R
f(z) dz = 0,
and if f is analytic on a disk containing a closed curve , then
_

f(z) dz = 0.
We can develop this idea further in two direction. One is to describe the most
general region such that if is a closed curve in and f is analytic on , then
_

f(z) dz = 0. Secondly, for a general C, we can describe closed curves


such that if f is analytic on , then
_

f(z) dz = 0.
The main issue is then to consider holes in the domain. This leads to study-
ing bounded components of the complement of . Additional ideas include
curves winding around points in the complement and continuous deformations
of curves.
5.46 Denition. A region C is simply connected if

C is connected.
5.47 Theorem. A region C is simply connected if and only if n(, a) = 0
for all closed curves and a

C .
Integration Theory 87
Proof. Suppose is simply connected. Take the component of

C that
contains , which also contains

C . But n(, a) is constant for a in a
component of

C and n(, a) = 0 at , so n(, a) 0 on

C .
For the converse, suppose is not simply connected, so

C is E

E,
where E

is the component containing , E

E = , E ,= , and both E
and E

are closed. Then the distance from E to E

is some > 0, and we can


take a E. Taking a grid of size at most /2 such that a is not on a grid line,
we see that for any square Q
j
in the grid,
n(Q
j
, a) =
_
_
_
0, a / Q
j
2i, a Q
j
.
Now take all Q
j
that do not intersect E

. This includes all squares that intersect


E, so the interior of the union of these squares is a neighborhood of E, and its
boundary is a nite union of curves contained in . Furthermore,
n(, a) =

j
n(Q
j
, a) = 2i.
5.48 Denition. Let be a region and let be a closed curve in . We say
is homologous to zero in if n(, a) = 0 for all a

C. In this case, we
write 0.
With this new terminology, we can rewrite the previous theorem about sim-
ple connectivity.
5.49 Theorem. A region C is simply connected if and only if all closed
curves in are homologous to zero.
In fact, the idea of homologous easily transfers to nite sums of closed
curves. If
1
, . . . ,
k
are closed curves, dene
n
_
k

j=1

j
, a
_
=
k

j=1
n(
j
, a).
So we can talk about
1
+ +
k
being homologous to zero in .
The idea of continuous deformation of curves is handled by the following
denition.
5.50 Denition. Suppose
0
and
1
are two curves with the same endpoints.
We say
0
is homotopic to
1
in if there exists a continuous function
88 General Form of Cauchys Theorem
(coming soon)
Figure 5.6: Notice that 1 is not homologous to zero. However, the curve
1 + 2 0.
: [0, 1]
2
so that (0, t) =
0
(t), (1, t) =
1
(t), and (s, 0) and (s, 1)
are constant.
5.51 Theorem. A region is simply connected if and only if any pair of curves
in with the same endpoints is homotopic in .
5.52 Denition. A closed curve is contractable in if it is homotopic to
a constant map in .
5.53 Theorem. A region is simply connected if and only if every closed curve
in is contractable in .
5.54 General Cauchys Theorem. Let be a region, f be analytic on ,
a closed curve contained in , and homologous to zero, or equivalently, is
contractable. Then
_

f(z) dz = 0.
5.55 General Cauchy Integral Formula. Suppose f is analytic in a region
and let be a curve in homologous to zero. Let a be a point not on .
Then
_

f(z)
z a
dz = n(, a)f(a).
5.56 Corollary. If f is analytic on a simply connected and is a closed
curve in , then
_

f(z) dz = 0.
5.57 Corollary. If f is analytic on a simply connected region , then f has
an antiderivative on .
Proof. Integrals are path independent.
5.58 Corollary. If f is analytic on a simply connected region and nonzero
on , then log(f) and f

, for C, are well-dened on .


Proof. The function f

/f is analytic on , so it has antiderivative


F(z) =
_
z
z0
f

(z)
f(z)
dz
Integration Theory 89
over some path in . So F = f

/f, which implies


d
dz
(f(z)e
F(z)
) = f

(z)e
F(z)
f(z)
f

(z)
f(z)
e
F(z)
= 0,
so f(z)e
F(z)
is constant and at z
0
it equals f(z
0
). Hence f(z) = f(z
0
)e
F(z)
.
This means
log(f(z)) = log(f(z
0
)) +F(z),
where we can choose any value of log on the right. Finally,
f(z)

= exp(log(f(z))),
so f

is well-dened.
We now give another characterization of Cauchys theorem. Indeed, let
C(C) be the set of continuous functions on C and dene T

: C(C) C by
f
_

f dz for some xed homologous to zero in . This is a bounded linear


functional. If T

vanishes on
1
za
: a / , then it vanishes on the set of all
analytic functions on .
CHAPTER 6
Residue Calculus
6.1 The Residue Theorem
Residue calculus is a method for doing integrals using Cauchys theorem (The-
orem 5.54). The idea is that f is analytic on except at a set of isolated
singularities, call them z
j
. Let be a closed curve in homologous to zero
(coming soon)
Figure 6.1: The motivational idea for residue calculus.
and such that z
j
is empty. Around each z
j
, make a small circle
j
of
radius r
j
, with r
j
small enough that D(z
j
, r
j
) ,
j

k
= for all j ,= k,
and
j
= . Now
:=

j
n(, z
j
)
j
is a cycle homologous to zero in z
j
. We have
n
_

j
n(, z
j
)
j
, a
_
= n(, a)

j
n(, z
j
)n(
j
, a).
If a = z
k
for some k, then n(
k
, a) = 2i and n(
j
, a) = 0 for all j ,= k. This
implies n( , z
k
) = 0 for all k, and also n( , z) = 0 for all z C because its
91
92 The Residue Theorem
true for all terms. Hence by Cauchys theorem,
_
e
f(z) dz = 0;
that is,
_

f(z) dz =

j
n(, z
j
)
1
2i
_
j
f(z) dz.
Observe
_
j
f(z) dz is independent of radius r
j
if r
j
is suciently small.
6.1 Denition. The residue of f at the isolated singularity z
j
, denoted by
Res(f, z
j
), is the unique number
Res(f, z
j
) =
1
2i
_
|zzj|=rj
f(z) dz
for all suciently small r
j
.
The way the residue is dened in [Ahl79] is the following. The residue of f
at the isolated singularity z
j
, denoted Res(f, z
j
), is the unique number so
f(z)
Res(f, z
j
)
2i(z z
j
)
has an antiderivative on 0 < [z z
j
[ < r
j
for all suciently small r
j
.
Residues at poles are easy to compute. (At essential singularities, forget
about it!) If f(z) has a pole at z
j
, then
f(z) =
n

=1
a

(z z
j
)

+g(z),
where g is analytic in a neighborhood of z
j
and its nonzero at z
j
, and
_
j
f(z) dz =
n

=1
a

_
j
(z z
j
)

dz = a
1
2i.
6.2 Theorem. Suppose f has a pole of order n at z = z
j
. Then
Res(f, z
j
) = coecient of (z z
j
)
1
in expansion of f at z
j
=
1
(n 1)!
lim
zzj
_
d
dz
_
n1
_
(z z
j
)
n
f(z)
_
.
Residue Calculus 93
6.3 Corollary. If f has a simple pole at z
0
, then
Res(f, z
0
) = lim
zz0
(z z
0
)f(z).
Proof. Setting n = 1 in the theorem gives the result.
6.4 Residue Theorem. Suppose f is analytic in except at isolated singu-
larities z
j
and is homologous to zero in . Then
_

f(z) dz =

j
n(, z
j
) Res(f, z
j
).
Note that the Cauchy integral formula is a special case. If f(z) is analytic
on , then
f(z)
za
has a simple pole at z = a with residue f(a). Another special
case was our method of counting zeros using
_

(z)
f(z)
dz = n(f(), 0)
because writing f(z) = (z z
j
)
nj
g(z) at a zero implies f

/f has a simple pole


with residue n
j
.
This begs the question: Can we also count poles this way? If f has a pole
at w
j
, then f(z) = (z w
j
)
mj
g(z) and
f

(z)
f(z)
=
m
j
z w
j
+
g

(z)
g(z)
.
So f has a simple pole with residue m
j
.
6.5 The Argument Principle. If f is meromorphic
1
in with poles w
j
of
orders m
j
, and if it has zeros at z
j
with multiplicities n
j
, then for homolo-
gous to zero in we have
n(f(), 0) =
_

(z)
f(z)
dz =

j
n
j
n(, z
j
)

j
m
j
n(, w
j
).
6.6 Rouches Theorem. If f and g are analytic in , is a simple closed
curve homologous to zero in , and [f(z) g(z)[ < [f(z)[ on , then f and g
have the same number of zeros in .
1
This means f is analytic in a region except at a discrete set of poles.
94 Calculating Integrals
Proof. We have

1
f(z)
g(z)

< 1
on and
g
f
: z : n(, z) = 2i : [1 [ < 1.
This implies n(
g
f
(), 0) = 0, so the number zeros of g/f is equal to the number
of poles of g/f. But the zeros of g/f are the zeros of g and the poles of g/f are
the zeros of f.
6.2 Calculating Integrals
Residues are used to evaluate various integrals. A typical feature of such in-
tegrals are the integrand is a restriction of an analytic function to a curve.
Another feature is by taking some extra curves, one can change the integral
into one along a curve homologous to zero (or a limit of such curves) bounding
a region where the integral is analytic except for some isolated singularities. One
can say something about the integrals along the new pieces of curves (frequently
what one can say is that they go to zero when we take an appropriate limit).
6.7 Example. We will evaluate
_

1
1 +x
2
dx.
By the denition of an improper Riemann integral, the integral equals
lim
r
lim
s
_
s
r
1
1 +x
2
dx.
Since the integral is in L
1
(R), the integral is
lim
R
_
R
R
1
1 +x
2
dx.
So we have an integrand which, on [R, R] R, coincides with
1
1+z
2
, which is
an analytic function except at i where it has simple poles.
Let be [R, R] [z[ = R : Im(z) > 0. So is homologous to zero in H,
Residue Calculus 95
and if R 0, in particular, if R > 1, then
_

1
1 +z
2
dz = 2i Res(f, i) = 2i lim
zi
(z i)f(z) = ,
since f(z) =
1
1+z
2
=
1
(z+i)(zi)
has a simple pole at i.
Now

_
R
R
1
1 +x
2
dx

_
|z|=R
zH
1
1 +z
2
dz

sup
|z|=R
zH
1
[1 +z
2
[
R.
But [1 +z
2
[ [z[
2
1 R
2
1, so
sup
|z|=R
zH
1
[1 +z
2
[
R
R
R
2
1
.
Hence
lim
R
_
R
R
1
1 +x
2
dx = .
A general type of example is
_

f(x) dx,
where f(x) is a rational function in C(x) with a zero of order at least two at .
6.8 Example. Evaluate the integral
_
2
0
1
(3 + cos )
2
d.
If we let z = e
i
, then dz = ie
i
d and
cos =
1
2
(e
i
+e
i
) =
1
2
_
z +
1
z
_
.
So
_
2
0
1
(3 + cos )
2
d =
_
|z|=1
1
(3 + (
1
2
(z +
1
z
))
2
dz
iz
=
_
|z|=1
4z dz
i(6z +z
2
+ 1)
2
.
96 Calculating Integrals
Now we nd the poles. If 6z +z
2
+ 1 = 0, then
z =
6

36 4
2
=
6

32
2
= 3

8.
Observe

is not in D, but

3 +

< 1. So
_
|z|=1
4z
(6z +z
2
+ 1)
2
dz
i
=
_
|z|=1
4z
(z (3 +

8))
2
(z (3

8))
2
. .
f(z)
dz
i
.
f(z) has a pole of order 2 at 3 +

8 in D. To compute the residue at this pole


we have
Res(f(z), 3 +

8) =
d
dz
(z (3 +

8))
2
f(z)

z=3+

8
=
3
16

2
.
So
_
|z|=1
4z dz
i(6z +z
2
+ 1)
2
=
2i
i
3
16

2
=
3
8

2
.
This example is the prototype for integrals of the form
_
2
0
f(cos , sin ) d,
where f is a rational function in C(x).
6.9 Example. Evaluate
_

cos x
x
2
+a
2
dx, a R.
This integral looks like
Re
_
_

e
iz
dz
z
2
+a
2
_
or
_

e
iz
+e
iz
2(z
2
+a
2
)
dz.
It is a little cleaner to use the integral on the left, so we proceed with that one.
We have
_

e
iz
dz
z
2
+a
2
= lim
R
_
R
R
e
iz
dz
z
2
+a
2
because

e
ix
x
2
+a
2

1
x
2
+a
2
Residue Calculus 97
is in L
1
(R). Add the semicircle [z[ = R with z H to get a curve homologous
to zero in the upper half-plane. So
_

e
iz
dz
z
2
+a
2
= 2i Res
_
e
iz
z
2
+a
2
, ia
_
.
The function has a simple pole at i[a[, so
Res
_
e
iz
z
2
+a
2
, ia
_
=
e
i(i|a|)
2i[a[
=
e
|a|
2i[a[
.
Hence
_

e
iz
dz
z
2
+a
2
=
2i
2i[a[
e
|a|
=
e
|a|
[a[
.
As previously, we see that

_
|z|=R
zH
e
iz
z
2
+a
2
dz

R
R
2
[a[
2
sup
|z|=R
zH
[e
iz
[
R
R
2
[a[
2
0
as R since
[e
iz
[ = e
Re(iz)
= e
Im(z)
1.
Thus
_

cos x
x
2
+a
2
dx =
e
|a|
[a[
, a R.
This example is the prototype for
_

e
ix
f(x) dx,
where f(x) is again a rational function with a zero of order at least two at
. Note that the e
ix
can be replaced by cos x or sin x by taking the real and
imaginary parts.
In fact, we can do better in this case. We can do the integral above where
f(x) has a zero of order one at . However, this case is a bit more fussy.
Consider the curve which is the rectangle [r, s] [0, t], as shown in Figure
6.2. Then
lim
r
lim
s
lim
t
_

e
ix
f(x) dx =

2i(Res H).
98 Calculating Integrals
Figure 6.2: Use this contour to integrate e
ix
f(x), where f(x) is a rational function
with a zero of order one at innity.
Let
2
= [r, s] t. Then

_
2
e
ix
f(x) dx

sup
2
[e
iz
f(z)[ (r +s)
since [f(z)[ c/[z[ as z and [e
iz
[ e
t
on
2
. So

_
2
e
ix
f(x) dx

(r +s)e
t
C,
where C is some constant. For
3
dened by r [0, t] we have
[R(z)[
c
[z[

c
r
and the length is t. Hence

_
3
e
ix
f(x) dx

Ct
r
,
and similarly

_
1
e
ix
f(x) dx

Ct
s
.
Set t = min(

r,

s). Then

_
1
+
_
3

r
+
C

s
0.
Residue Calculus 99
Thus
_
0
+
_
1
+
_
2
+
_
3
=

residues.
But sending t gives
_
2
0,
so

_
0

residues

_
1

_
3

C
_
1
r
+
1
s
_
0
as r, s .
6.10 Example. Evaluate
_

e
ax
1 +e
x
dx, 0 < a < 1.
Its clear we can equate the integral to
_

e
az
1 +e
z
dz.
Since the function is integrable, we have
_

e
az
1 +e
z
dz = lim
R
_
R
R
e
az
1 +e
z
dz.
But taking the usual semicircle of radius R in the upper half-plane causes prob-
lems. However, using the contour given by the boundary of [R, R] [0, 2i]
allows us to use the periodicity of the exponential. Let
j
, j = 0, 1, 2, 3, be
the line segments of the contour, starting with
0
= [R, R] R and continue
counterclockwise. Then
_
0
+
_
1
+
_
2
+
_
3
= 2i Res
_
e
az
1 +e
z
, i
_
,
because at e
z
= 1, we have [e
z
[ = e
x
= 1, which implies x = 0, so z = iy. Then
e
iy
= 1 means y = i. The pole z = i is a simple pole because e
z
+ 1 = 0
100 Calculating Integrals
and the derivative of e
z
+ 1 is nonzero at z = i. So the residue is given by
Res
_
e
az
1 +e
z
, i
_
= lim
zi
(z i)e
az
1 +e
z
= e
ai
lim
zi
z i
1 +e
z
= e
ai
1
1
= e
ai
,
by lHopitals rule.
Now on
1
, the integrand is

e
a(R+iy)
1 +e
R+iy

e
aR
e
R
1
.
So

_
1
e
az
1 +e
z
dz

e
aR
e
R
1
2 0
as R . Similarly, for
3
, the integrand has a bound

e
R+iy
1 +e
R+iy

e
aR
1 e
R
.
Hence

_
3
e
az
1 +e
z
dz

2e
aR
1 e
R
0
as R . Then
_
2
e
az
1 +e
z
dz =
_
R
R
e
a(x+2i)
1 +e
x+2i
dx =
_
R
R
e
a2i
e
ax
1 +e
x
= e
a2i
_
0
e
ax
1 +e
x
dx.
So
_
0
+
_
1
+
_
2
+
_
3
= 2i(e
ai
)
and

(1 e
2ai
)
_
R
R
e
ax
1 +e
x
dx + 2ie
ai

_
1

_
3

0
as R . Hence
_
R
R
e
ax
1 +e
ax
dx =
2ie
ai
1 e
2ai
=
2i
e
ai
+e
ai
=

sin a
.
Residue Calculus 101
6.11 Example. Evaluate
_

0
x
1/4
x
2
+ 9
dx.
The function z
1/4
is not analytic on all of C, so we need to cut out (for example)
a ray from 0 to . Intuitively, we have the picture as in Figure 6.3. Then we
have the branch of z
1/4
on C R
0
. Then
(coming soon)
Figure 6.3: This is a contour deleting the origin and the positive real axis.

R
z
1/4
z
2
+ 9
dz

2R
R
1/4
R
2
9
0
as R . Also

z
1/4
z
2
+ 9
dz

2

1/4
9
2
0
as 0. So only
+
and

are going to contribute the integral. We have


_
+
z
1/4
z
2
+ 9
dz =
_
R

x
1/4
x
2
+ 9
dx
and
_

z
1/4
z
2
+ 9
dz =
_

R
e
i/2
x
1/4
x
2
+ 9
dx = i
_
R

x
1/4
x
2
+ 9
dx.
We want to say
_
++
R
++
= 2i
_
Res
_
z
1/4
z
2
+ 9
, 3i
_
+ Res
_
z
1/4
z
2
+ 9
, 3i
__
= 2i
_
(3i)
1/4
6i
+
(3i)
1/4
6i
_
=

3
_
(3i)
1/4
(3i)
1/4
_
=

3
_
3
1/4
e
i/8
3
1/4
e
3i/8
_
,
since z
1/4
= [z[
1/4
e
i/4
where z = [z[e
i
. Taking R and 0,
(1 i)
_

0
x
1/4
x
2
+ 9
dx =
3
1/4
3
_
e
i/8
e
3i/8
_
.
102 Calculating Integrals
Thus
_

0
x
1/4
x
2
+ 9
dx = 3
3/4
e
i/8
e
i3/8
1 i
=
3
3/4
2
(e
i/8
+e
5i/8
+e
5i/8
+e
i/8
)
=
3
3/4
2
_
cos

8
+ 2 cos
5
8
_
= 3
3/4
_
cos

8
+ cos
5
8
_
.
With this previous example as a prototype, we can do any integral of the
form
_

0
x

f(x) dx,
where f is a rational function which has a zero of order at least two at and
0 < < 1. We can also evaluate
_

0
ln x
x
2
+ 1
dx,
or
_

0
ln(x)f(x
2
) dx,
f a rational function, in this form (almost!).
CHAPTER 7
Harmonic Functions
7.1 Harmonic Conjugates
A function u: R, where C, is said to be harmonic if u C
2
and
u =
_

2
x
2
+

2
y
2
_
u = 0
on . Equivalently, we can dene a harmonic function u: U R, for U R
2
.
We can make several observations about harmonic functions.
is a linear dierential operator, so linear combinations of harmonic
functions are harmonic.
Linear functions are harmonic globally. The function log r is harmonic
away from 0, where r =
_
x
2
+y
2
.
If f = u + iv is analytic on , then u and v are harmonic on by the
Cauchy-Riemann equations. Indeed, u
x
= v
y
and u
y
= v
x
implies
u
xx
+u
yy
= v
xx
+v
yy
= 0.
If f is analytic and has no zeros on , then log [f[ is harmonic on .
If u is harmonic, then u
x
iu
y
is analytic.
103
104 Harmonic Conjugates
A somewhat harder question to answer: if u is a harmonic function on ,
does there exist v : R harmonic such that f = u +iv is analytic?
7.1 Denition. If u is harmonic on and there exists u such that u + i u is
analytic on , then u is called the harmonic conjugate of u.
We want the Cauchy-Riemann equations to hold; i.e., we want u
x
= u
y
and
u
y
= u
x
. So we try to use that information to dene u. What we have is
d u = u
x
dx + u
y
dy = u
y
dx +u
x
dy.
Ten we can try to dene u by taking a basepoint (x
0
, y
0
) and setting
u =
(x,y)
_
(x0,y0)
d u.
We know u is well-dened if and only if
_

(u
y
dx +u
x
dy) = 0
for all closed curves . Greens theorem (Theorem 5.5) would tell you
_

(u
y
dx +u
x
dy) =
__
inside
(u) dA.
(This is only meaningful for elementary curves like rectangles or circles.) To
prove this way, we would need to repeat lots of the work from the proof of
Cauchys theorem. Instead, we prove by reducing to Cauchys theorem.
7.2 Lemma. If u is harmonic in and is homologous to zero in , then
_

(u
y
dx +u
x
dy) = 0.
Proof. Let g = u
x
iu
y
. Then g is analytic on by the remark above (whose
Harmonic Functions 105
proof uses the Cauchy-Riemann equations). So
0 =
_

g dz =
_

(u
x
iu
y
)(dx +i dy)
=
_

(u
x
dx +u
y
dy) +i
_

(u
y
dx +u
x
dy)
=
_

du +i
_

d u.
Both real and imaginary parts are zero, so the result follows.
From this lemma, we conclude the following.
7.3 Theorem. If u is harmonic in a simply connected region , then u has
harmonic conjugate in .
We should note that if is not simply connected, then there exist harmonic
functions u on such that the natural harmonic conjugate is multi-valued.
What will happen is that u will not be well-dened because
_

d u = Im
__

g(z) dz
_
= 2m, m Z,
by imitating the proof of the lemma.
7.4 Example. The quintessential example is u(z) = log [z[, which is harmonic
on C

. For any simply connected region C

, there exists u = arg(z) such


that u + i u = log z is analytic. But for not homologous to zero in C

,
_

d u
will be a multiple of 2, so u ends up multi-valued and u +i u is multi-valued.
7.2 Harmonic Functions and Normal Derivatives
Recall that if u is harmonic and is homologous to zero, then
_

(u
y
dx +u
x
dy) = 0.
If is dierentiable at a point z, then dz (along ) is [dz[e
i
for some angle
= (z). Therefore dx = [dz[ cos and dy = [dz[ sin , so
u
y
dx +u
x
dy = (u
x
, u
y
) (sin , cos ) [dz[ = u n [dz[ =
u
n
[dz[,
106 Harmonic Functions and Normal Derivatives
where is the usual dot product on R
2
, n is the unit normal vector, and the
partial derivative on the right is the directional derivative of u in the direction
of n normal to .
(coming soon)
Figure 7.1: A picture of the normal derivative
7.5 Lemma. For any homologous to zero in ,
_

(u
y
dx +u
x
dy) = 0.
This follows from the fact that
_

u
n
[dz[ = 0
for all homologous to zero in .
This is useful when looking at circle, because the normal direction is easy to
get. If u is harmonic in a disk around zero, then
u
n
=
u
r
, r =
_
x
2
+y
2
.
We can conclude the following.
7.6 Lemma. If u is harmonic in a disk around zero, then
_
2
0
r
u
r
d = 0
for all suciently small r.
7.7 Lemma. If u is harmonic in an annulus r
1
< [z[ < r
2
around zero, then
_
2
0
r
u
r
d
is independent of r (r
1
, r
2
).
Proof. If r and r

are in (r
1
, r
2
), then let
r
= z : [z[ = r and
r
= z : [z[ =
r

.
Harmonic Functions 107
(coming soon)
Figure 7.2: The annuli in the proof of the lemma.
The cycle
r

r
is homologous to zero in the annulus. So
_
r
r

r
u
r
d = 0
as needed.
There is a generalization of the fact
_

u
n
[dz[ = 0
for harmonic u in and homologous to zero in .
7.8 Gauss-Green Theorem. We have
_

_
u
v
n
v
u
n
_
[dz[ =
__
inside
(u v v u) dA.
We will only need the weaker version:
7.9 Theorem. If u and v are harmonic on and is homologous to zero in
, then
_

_
u
v
n
v
u
n
_
[dz[ = 0.
Proof. There are several ways to prove this. One could prove for a rectangle
and then transfer to by using a ne enough grid. The proof for a rectangle
uses either Cauchys theorem or by taking a ne subdivision of the rectangle and
proof locally using the linear expansion (like our proof of Greens theorem). The
proof in [Ahl79] uses Cauchys theorem, and it uses the fact that the rectangle
is simply connected, so
v
n
[dz[ = d v,
u
n
[dz[ = d u,
where u and v are harmonic conjugates of u and v. Thus
_
u
v
n
v
u
n
_
[dz[ = u d v v d u = exact dierential +F dz,
for some analytic F. See [Ahl79, p. 164] for full details.
108 Harmonic Functions and Normal Derivatives
Applying this theorem to a circle, we have u and v harmonic in the annulus
between r
1
and r
2
. Then
_
C(a,r)
_
u
v
n
v
u
n
_
[dz[
is independent of r (r
1
, r
2
), so we have
_
2
0
_
u
v
r
v
u
r
_
r d.
In particular, if we take v = log r, we get
_
2
0
u r log r
u
r
d
is independent of r (r
1
, r
2
), so it equals some constant c
1
. Recall we had
_
2
0
r
u
r
d
is independent of r (r
1
, r
2
), so it equals c
2
C. This implies
_
2
0
u d = c
1
+c
2
log r, c
1
, c
2
C.
7.10 Theorem. If u is harmonic in r
1
[z[ r
2
, then
_
2
0
u(re
i
) d = c
1
+c
2
log [z[
in the annulus.
7.11 Corollary. If u is harmonic in [z[ R, then
_
2
0
u(re
i
) d = 2u(0).
Proof. We have

u
r

|u(0)|(1 +)
Harmonic Functions 109
in some neighborhood of 0. So
_
2
0
r log r
u
r
d 0
as r 0. Hence
_
2
0
u d = c
1
,
but r suciently small implies

_
2
0
u(re
i
) d
_
2
0
u(0) d

2 max
|z|=R
[u(z) u(0)[ 0,
so c
1
= 2u(0).
7.12 Mean Value Property for Harmonic Functions. If u is harmonic on
[z a[ < r and continuous on [z a[ r, then
u(a) =
1
2
_
2
0
u(a +se
i
) d, s [0, r].
Proof. Apply translate of the previous corollary to any disk [z a[ r

< r and
take the limit as r

r. (Note that u(a + r

e
i
) u(a + re
i
) uniformly in
as r

r because u is uniformly continuous on [z a[ r.)


7.13 Maximum Principle for Harmonic Functions. If u is harmonic and
nonconstant on a region , then u does not attain a maximum or minimum on
. If u is harmonic on a compact E, then the maximum and minimum occur
on E (and only on E unless u is constant).
Proof. Suppose u has maximum value at a . Then for all r suciently small,
u(a +re
i
) u(a) =
1
2
_
2
0
u(a +re
i
) d.
If there exists a point with u(a + re
i
) < u(a), then the inequality holds on a
neighborhood of this point, so
1
2
_
2
0
u(a +r
i
) d < u(a),
which is a contradiction. So we conclude that u u(a) in a neighborhood of
a. So then z : u(z) = u(a) is closed because u is continuous and it is open
110 Harmonic Functions and Normal Derivatives
because just derived u u(a) in a neighborhood of where u(a) is achieved.
Connectivity implies u u(a) on .
7.14 Example. We have the following consequence in functional analysis. Con-
sider the situation where u is harmonic on the bounded region and u = f on
, where f is a continuous function. This is a solution to a Dirichlet problem:
nd u such that u = 0 on , and u is a prescribed continuous function on
the boundary . The maximum principle says |u| |f|, where | | is the
sup norm, and the map f u(z) is a bounded linear functional C() R,
for each xed z . So if you can solve u = 0 on and u = f on for
f C(), then the map f u(z) is a bounded linear functional. So the Riesz
representation theorem says the map is
f
_

f() d
z
()
for some Borel measure
z
on . The mean value property is an example of
this. Indeed
u(a) =
1
2
_
2
0
u(a +re
i
) d,
where u(a + re
i
) is the values on the boundary of the disk, d is a measure
on the boundary of the disk, and u(a) is the function value at the center of the
disk.
We can nd more by moving the point by a fractional linear transformation.
7.15 Theorem. If F is analytic on and u is harmonic on F(), then u F
is harmonic on .
Proof. Take z and take a small disk around F(z) so small that u is harmonic
on the disk. This disk is simply connected, so u has a harmonic conjugate u on
the disk and G := u +i u is analytic on the disk. But then G F is analytic at
z, so u F = Re(G F) is harmonic at z.
7.16 Theorem. Let u be harmonic on the closed unit disk D. Then
u(z) =
1
2
_
2
0
u(e
i
)
1 [z[
2
[e
i
z[
2
d, z D.
Proof. Let F
z
: D D be the linear fractional transformation by
F
z
() =
+z
1 +z
.
Harmonic Functions 111
Since (u F
z
)() is harmonic on D, we have
u(z) = (u F
z
)(0) =
1
2
_
2
0
(u F
z
)(e
i
) d.
The map F
z
takes the unit circle to itself, so we can make the change of variables
F
z
(e
i
) = e
i
. Then
F

z
(e
i
)e
i
i d = e
i
i d,
and
F

z
(F
1
z
(e
i
))F
1
z
(e
i
) d = e
i
d.
We can compute the left side from F
z
F
1
z
= id, so (F

z
F
1
z
) (F
1
z
)

= 1.
Hence
d =
(F
1
z
)

(e
i
)
F
1
z
(e
i
)
e
i
d.
But
F
z
() = w =
+z
1 +z
implies
=
w z
1 zw
,
and so
F
1
z
(w) =
w z
1 wz
.
So
(F
1
z
)

=
1
1 zw

(w z)(z)
(1 zw)
2
and
(F
1
z
)

(w)
F
1
z
(w)
=
(1 zw) +z(w z)
(1 zw)
2

1 zw
w z
=
1 [z[
2
(1 zw)(w z)
.
Substituting w = e
i
and multiplying both sides by e
i
gives the desired result.
Note the case when u 1, which is a harmonic function. We have
1
2
_
2
0
1 [z[
2
[e
i
z[
2
d = 1, z D.
The integrand is the representing measure in the Riesz representation theorem
for the bounded linear functional u[
D
u(z).
112 Harmonic Functions and Normal Derivatives
7.17 Denition. The function
P
z
() =
1
2
_
1 [z[
2
[e
i
z[
2
_
= Re
_
e
i
+z
e
i
z
_
is called the Poisson kernel (at z) for the unit disk.
We already saw
_
2
0
P
z
() d = 1,
and P
z
() 0. You can easily see that if [[ = 1 and look at P
z
() as z ,
then P
z
() gets concentrated at . The limiting object (in the sense of
distibutions) is a Dirac mass at .
CHAPTER 8
Series and Analytic Functions
8.1 Relationship Between Series and Analytic
Functions
If you have a sequence of analytic functions f
j
on a region , what conditions
or type of convergence would imply f = limf
j
is analytic on ? If we take
z and r < dist(z, ), then
f
j
(z) =
1
2i
_
|z|=r
f
j
()
z
d. (8.1)
If f is analytic, then
f(z) =
1
2i
_
|z|=r
f()
z
d. (8.2)
A sucient condition for (8.1) to imply (8.2) is that f
j
f uniformly on
[ z[ = r. Also, if f
j
f uniformly on [ z[ r, then
0 =
_

f
j
d
_

f d
113
114 Relationship Between Series and Analytic Functions
for any closed curve in [ z[ < r, so f will be analytic on [ z[ < r by
Moreras theorem. So we have determined that in order to have f = limf
j
analytic on , it suces that for every z , there exists a disk : [ z[
r on which f
j
f uniformly.
8.1 Theorem (Weierstrass). If f
j
is a sequence of analytic functions on
and f
j
f uniformly on all compact subsets of , then f is analytic on .
Moreover, f
(k)
j
f
(k)
uniformly on compact subsets of .
To get that the derivatives converge, use
f
(k)
j
(z) =
k!
2i
_
|z|=r
f
j
()
( z)
k+1
d. (8.3)
The quantity in (8.4) converges to
k!
2i
_
|z|=r
f()
( z)
k+1
d = f
(k)
(z). (8.4)
Proof. Use Moreras theorem on disks and the above paragraph.
8.2 Corollary. If f
j
is analytic on and

j0
f
j
converges uniformly on
compact subsets of , then f =

j0
f
j
is analytic on and is dierentiable
term-by-term.
Warning: We know power series are dierentiable term-by-term on the disk
of convergence, but, in general, series might not be dierentiable term-by-term.
For example, consider the series

j=1
cos(2
j
x)
j
2
, x R.
This series converges no problem, but the series of derivatives doesnt converge
because you get a 2
j
in the numerator after just one derivative.
8.3 Corollary. If f
j
is a sequence of analytic functions on , K

is a
sequence of compact sets in so

= , and f
j
f uniformly on each
K

, then f is analytic on .
8.4 Hurwitz Theorem. If f
j
is a sequence of analytic functions on ,
f
j
(z) ,= 0 on for every j, and f
j
f uniformly on compact sets in , then
either f 0 on or f(z) ,= 0 on .
Series and Analytic Functions 115
Proof. The argument principle says that if you have a circle in then the
number of zeros of f
j
inside is equal to
1
2i
_

j
(z)
f
j
(z)
dz,
and the number of zeros of f inside is
1
2i
_

(z)
f(z)
dz, f(z) ,= 0 on .
The idea is that the rst integral should converge to the second integral provided
f
j
stays away from 0 on .
Suppose now f(z
0
) = 0 and f , 0. Then z
0
is an isolated zero, so there
exists = : [ z
0
[ = r such that f() ,= 0 on , so [f()[ c > 0 on .
Then f
j
f uniformly on , which implies there exists J such that [f
j
()[
c
2
on for every j J. Furthermore,
min

jJ
[f
j
()[ = d > 0
because f
j
, 0 on . Then
sup

j
()
f
j
()

f

()
f()

2
min(c, d)
sup

j
() f

()

0.
So
0 =
1
2i
_

j
()
f
j
()
d
1
2i
_

()
f()
d ,= 0,
which is a contradiction.
8.2 Taylor Series Developments
We proved earlier in the course that if f is an analytic function on [z a[ r,
then we have the Taylor polynomials
F
n
(z) =
n

j=0
f
(j)
(a)
j!
(z a)
j
116 Taylor Series Developments
and f(z) = F
n
(z) +f
n+1
(z)(z a)
n+1
, where
f
n+1
(z) =
1
2i
_
|a|=r
f()
( a)
n+1
( z)
d,
and f
n+1
(z)(z a)
n+1
0 uniformly on compact subsets of [z a[ < r (which
implies F
n
(z) f(z) uniformly on compact disks). Moreover,
f(z) =

j0
f
(j)
(a)
j!
(z a)
j
, on [z a[ < r.
8.5 Corollary. If f is analytic on , a , and r = dist(a, ), then
f(z) =

j0
f
(j)
(a)
j!
(z a)
j
on compact subsets of [z a[ < r.
Usually we get series by termwise dierentiation and integration of known
series. For example, the series expansions of e
z
and
1
1z
are useful. We can also
use partial fractions and linear transformations of the geometric series.
As important, one needs to able to manipulate nite order expressions.
It is easy to add and subtract Taylor series; we have

n0
a
n
z
n
+

n0
b
n
z
n
=

n0
(a
n
+b
n
)z
n
.
Multiplying is bit more tricky:
_

n0
a
n
z
n
__

n0
b
n
z
n
_
=

n0
_
n

m=0
a
m
b
nm
_
z
n
.
Its useful to think of

n0
a
n
z
n
as
F
N
(z)
N

n=0
a
n
z
n
plus [z
N+1
] = stu divisible by z
N+1
. This helps us to divide Taylor series.
Suppose we have

a
n
z
n

b
n
z
n
, b
0
,= 0
Series and Analytic Functions 117
and we want the rst N coecients. Let
P
N
=
N

n=0
a
n
z
n
, Q
N
=
N

n=0
b
n
z
n
.
Then
P
N
(z)
Q
N
(z)
= R
N
(z) + [z
N+1
]
by long division.
8.6 Example. Consider
z +z
2
2z
3
4 +z 2z
2
+z
3
.
This is
0 +
1
4
z +
3
16
z
2

11
64
z
3
+ [z
4
]
because
z
4
(4 +z 2z
2
+z
3
) = z +
z
2
4

z
3
2
+ [z
4
]
=
3
16
z
2
(4 +z + )
=
3
4
z
2
+
3
16
z
3
+ [z
4
]
11
64
z
3
(4 + )
=
11
16
z
3
+ [z
4
].
Now we consider composition of series. Let f(w) =

n0
a
n
w
n
and g(z) =

n0
b
n
z
n
. What is (f g)(z)? If b
0
,= 0 then were screwed (because its
hard). If b
0
= 0 then observe [w
N+1
] = [(g(z))
N+1
] = [z
N+1
]. We conclude that
(f g)(z) = P
N
(Q
N
(z)) + [z
N+1
].
The denition of composition allows us to consider inverses. We want (f
g)(z) = z (up to nite order), where b
0
= 0 and b
1
,= 0. So can solve as
P
N
(Q
N
(z)) = z + [z
N+1
]
and iterate over N to get the coecients.
118 Laurent Series
8.3 Laurent Series
Recall again the theorem which states if f is analytic on and a , and take
r such that z : [z a[ < r , then
f(z) =

j0
f
(j)
(a)
j!
(z a)
j
,
in the sense of uniform convergence on compact subsets of D(a, r). We can
move this around via a linear fractional transformation.
8.7 Example. Suppose f is analytic in the complement of the disk D(a, r).
One can map (z) =
1
za
to take the complement of D(a, r) to D((a),
1
r
) to
get
f(z) =

j0

j
_
1
z a
_
j
,
in the sense of uniform convergence on compact subsets (of

C) in

C D(a, r).
So
f(z) =

j0

j
(z a)
j
denes an analytic function outside a disk D(a, r), and
g(z) =

j0

j
(z a)
j
denes an analytic function inside a disk D(a, R). If r < R, then f +g is analytic
on the annulus A(a, r, R) and f +g has series expansion
(f +g)(z) =

jZ
b
j
(z a)
j
with
b
j
=
_

j
, j > 0,

j
, j < 0,

0
+
0
, j = 0.
(8.5)
This has a converse obtained just by using Cauchys theorem. Suppose h
is analytic on an annulus A(a, r, R). Let
1
and
2
be as in Figure 8.1 and let
Series and Analytic Functions 119
z A(a, r, R). Cauchys theorem says
h(z) =
1
2i
__
1
h() d
z

_
2
h() d
z
_
,
provided
1
and
2
are close enough to [ a[ = R and [ a[ = r, respectively.
Let g(z) be the rst integral and let the second integral be f(z). So f(z) is
(coming soon)
Figure 8.1: Cauchys theorem allows for a converse of the Laurent series
development.
analytic o
2
and g(z) is analytic o
1
. In particular,
f(z) =
1
2i
_
2
h() d
z
is analytic on z : [z a[ > r (by taking
2
close enough to z : [z a[ = r)
and
g(z) =
1
2i
_
1
h() d
z
gives a function analytic on D(a, R). Then f(z) is analytic outside z : [z a[ >
r, so there exists a series

j0

j
(z a)
j
= f(z)
(but we can show
0
= 0). Similarly,
g(z) =

j0

j
(z a)
j
on D(a, R). So h = f +g on A(a, r, R) gives
h(z) =

jZ
b
j
(z a)
j
,
where b
j
is as in (8.5). We have proved the following:
8.8 Theorem. Suppose f is analytic on A(a, r, R). Then there exists an ex-
pansion of f by
f(z) =

jZ
b
j
(z a)
j
.
120 Laurent Series
This is called a Laurent series
One nice consequence is that if you have a function which is analytic except
at a discrete set of singularities, then you can give series expansions except on
a collection of circles.
8.9 Example. Suppose f is analytic, except at z
1
= 1+i, z
2
= 2, and z
3
= 3i.
So there is a Taylor series expansion of f in D(0,

2). We also get a Laurent


series in A(0,

2, 2). There is also another expansion in the annulus A(0, 2, 3).


CHAPTER 9
Uniform Convergence
9.1 On Compact Subsets of
This the correct notion of convergence of analytic functions in that analyticity
is preserved in the limit. The natural question then is there a natural metric
corresponding to this topology? The answer is yes! We can get such a metric
as follows:
1. Take a compact exhaustion of ; that is, a sequence K

of compact sets
with K

for all and


= . Observe that if K is compact,


then there exists such that K

K. (Without loss of generality, we can


take K

K
+1
for all .)
2. Take a distance between functions on K

. First, take
d(z, w) =
[z w[
1 +[z w[
to be a new metric on C. Indeed, this is a metric on C and it is bounded
by 1. Then for f and g be continuous functions on , let

(f, g) = sup
zK

d(f(z), g(z)).
This is bounded by 1 since d(f(z), g(z)) 1.
121
122 On Compact Subsets of
3. We have a countable collection

(f, g) of distances between functions on


K

. Dene
(f, g) =

1
2

(f, g).
9.1 Lemma. The function : C() C() R
0
is a metric.
Proof. Symmetric and positivity is obvious. The triangle inequality is true on d
(left as an exercise), so it is true for

and thus true for . Finally, (f, g) = 0


if and only if

(f, g) = 0 for all if and only if d(f(z), g(z)) = 0 for all z K

and all if and only if f(z) = g(z) for all z

= if and only if f = g on
.
9.2 Theorem. Let f
j
and f be elements of C() for all j. Then (f
j
, f) 0
if and only if f
j
f uniformly on compact sets in .
Proof. If (f
j
, f) 0, then for all > 0 there exists J such that (f
j
, f) < for
all j J, which implies 2

(f
j
, f) < for all . So this implies

(f
j
, f) 0
for all . Take any compact set K. Then there exists such that K

K. Also,

(f
j
, f) 0 implies f
j
f uniformly on K

K with respect to d. So then


given

> 0, there exists J such that


sup
zK
d(f
j
(z), f(z)) <

for all j J. If

<
1
2
, then
d(z, w) =
[z w[
1 +[z w[
so d(z, w) <

implies [z w[ < 2

and
sup
zK
[f
j
(z) f(z)[ < 2

.
Conversely, suppose that f
j
f uniformly on compact sets in . Then
sup
zK

[f
j
(z) f(z)[ 0,
which implies
sup
zK

d(f
j
(z), f(z)) 0
Uniform Convergence 123
and so

(f
j
, f) 0 for all as j . Given > 0, take
0
such that

0
2

<

2
.
Then

0
2

(f
j
, f) <

2
.
For each <
0
, take J

such that j J

implies

(f
j
, f) <

2
. Then for
j J = max
0
J

,
we have
(f
j
, f) =
01

=1
2

(f
j
, f) +

0
2

(f
j
, f) <

2
+

2
= .
9.3 Theorem. The metric space (C(), ) is complete.
Proof. Take a sequence f
j
that is -Cauchy. Then if z , there exists K

z
so f
j
(z) is Cauchy. So we can dene a pointwise limit f(z) = lim
j
f
j
(z). If
K is compact, then K is contained in some K

and f
j
(z) is uniformly
Cauchy on K

, so on K: for every > 0 there exists J such that for all j, m J


we have
sup
zK

[f
j
(z) f
m
(z)[ < .
Then send m to get
sup
zK
[f
j
(z) f(z)[ < ;
that is, f
j
f uniformly on K. Hence f C() and (f
j
, f) 0.
9.4 Remark. For a point z , consider a disk U = D(z, r), for some r > 0,
contained in . Then the above argument implies that f
j
f uniformly on
the closure of U, and is hence uniform at z.
9.2 Normal Families
We are in a complete metric space (C(), ), so a subset T C() is compact if
and only if T is closed and totally bounded. The Bolzano-Weierstrass theorem
124 Normal Families
says T is compact if and only if every sequence f
j
in T has a convergent
subsequence with limit in the set. Recall that a set is precompact (or relatively
compact) if its closure is compact.
9.5 Denition. A family of functions T C() is called normal if T is
precompact in the metric; i.e., if every sequence f
j
in T has a subsequence
f
j
k
that converges uniformly on compact subsets of .
This notion is useful in large part because there is a theorem giving necessary
and sucient conditions for T to be normal (the Arzela-Ascoli theorem). (This
theorem will be even simpler for families of analytic functions. Well see why
later.)
Obviously T C() is normal if and only if it is -totally bounded. We
can convert this to a statement about total boundedness in the sup norms on
compact subsets of .
9.6 Theorem. The family T C() is normal if and only if for every compact
subset K , the set f[
K
: f T is totally bounded in the sup norm on K,
if and only if for every compact K and > 0, there exists f
1
, . . . , f
n
T
such that for every f T there exists j with
sup
zK
[f(z) f
j
(z)[ < .
Proof. Suppose K is compact and take > 0. Then there exists K

containing K. Since f is -totally bounded, it can be covered by a nite


number of

-balls. (We will nd

later.) That is, there exists f


1
, . . . , f
n
such that for every f T there exists j with (f, f
j
) <

. This implies

(f, f
j
) < 2

, which is equivalent to
sup
zK

[f(z) f
j
(z)[
1 +[f(z) f
j
(z)[
< 2

.
We can rewrite this as
sup
zK

[f(z) f
j
(z)[ <
2

1 2

< .
Then solve for

in terms of to get what is desired. So f[


K
: f T is totally
bounded in the sup norm on K K

.
Conversely, take > 0. We need to show that we can cover T by a nite
Uniform Convergence 125
number -balls in the -metric. Take
0
such that 2
0
< /2. Then
(f, g)
0

=1
2

(f, g) +

2
.
We are assuming that for any compact K , the set f[
K
: f T is totally
bounded in the sup norm on K. In particular, for K
0
(which is compact in )
and

, there exists f
1
, . . . , f
n
such that for every f T there exists j with
sup
zK

0
[f(z) f
j
(z)[ <

.
This implies
sup
zK

0
[f(z) f
j
(z)[
1 +[f(z) f
j
(z)[
<

2
,
for a suitable choice of

. (Without loss of generality, the K

s are nested, so
the same is true for all K

with
0
.) Hence

(f, f
j
) <

2
,
0
,
which yields
(f, f
j
)
0

=1
2

2
+2 .
9.7 Lemma. If T C() is normal, then E := f(z) : f T is precompact
for every z . That is, the set is bounded.
Proof. Since T is normal, any sequence f
n
(z) of points from E has a uni-
formly convergence subsequence f
nj
on a compact subset of ; i.e., if f
nj
(z)
converges. So sequences f
n
(z) from E have convergence subsequence, which
implies E is precompact (in C).
9.8 Denition. A family T is called equicontinuous on a set K if for every
> 0 there exists > 0 such that [f(z) f(w)[ < for all z, w K with
[z w[ < .
9.9 Lemma. If T is normal on , then T is equicontinuous on any compact
K .
Proof. Since T is normal, for every > 0 there exists f
1
, . . . , f
n
such that for
126 Normal Families
all f T there exists j with
sup
zK
[f(z) f
j
(z)[ <

3
.
Thus
[f(z)f(w)[ [f(z)f
j
(z)[+[f
j
(z)f
j
(w)[+[f
j
(y)f(y)[
2
3
+[f
j
(z)f
j
(w)[.
Since each f
j
is continuous, it is uniformly continuous on the compact set K,
so there exists
j
such that [f
j
(z) f
j
(w)[ <

3
whenever [z w[ <
j
for all
z, w K. Let = min(
1
, . . . ,
n
). The proof from here is easy and is left to
the reader.
9.10 Arzela-Ascoli Theorem. The collection T C() is a normal family
if and if
(1) T is equicontinuous on every compact K , and
(2) for every z , the set f(z) : f T is precompact.
Proof. The (=) direction is the lemmas above.
To prove the other direction, take a dense subset z
r
in and take a se-
quence f
n
from T. First, we need to build a subsequence converging at points
z
r
. Iteratively, the sequence f
n
(z
1
) is bounded, so there exists a convergent
subsequence; i.e., the sequence f
n1,m
(z
1
) converges. The sequence f
n1,m
is
increasing. Now f
n1,m
(z
2
) is bounded, so it has a convergent subsequence
f
n2,m
, and so on. Inductively, there exists a subsequence f
nt,m
such that
f
nt,m
converges at z
1
, . . . , z
t
. The diagonal sequence f
nm,m
converges at all
points in a dense set.
Now we use equicontinuity to see f
nm,m
converges uniformly on compact
sets. Let K be a compact set and take > 0. Take > 0 so that [x y[ <
implies [f(x) f(y)[ < for all x, y K and f T. Cover K by a nite
number of -balls with centers at points z
r
from the dense set. Then
[f
nm,m
(x) f
np,p
(x)[ [f
nm,m
(x) f
nm,m
(z
r
)[
+[f
nm,m
(z
r
) f
np,p
(z
r
)[ +[f
np,p
(z
r
) f
np,p
(x)[.
But we can take z
r
within of x, so the latter is at most
2 +[f
nm,m
(z
r
) f
np,p
(z
r
)[.
Uniform Convergence 127
If m and p are suciently large, then convergence at z
r
implies that quantity
above is at most 3. So given > 0, we can take M so large that m, p > M,
which implies
[f
nm,m
(x) f
np,p
(x)[ <
for all x K. So the subsequence f
nm,m
is uniform Cauchy on K. So we can
cover K by a nite number of -balls with centers at points z
r
from the dense
set.
9.11 Montels Theorem. If T is a family of analytic functions on , then T
is normal if and only if T is uniformly bounded on every compact subset K of
. That is, if and only if for every K there exists M
K
such that
sup
zK
fF
f(z) M
K
.
Proof. If T is normal and K is compact in , then equicontinuity of T implies
there exists > 0 so that f T varies by some value less than 1 on -balls;
that is, if [x y[ < for x, y K, then [f(x) f(y)[ < 1. But K can be
covered by a nite number of -balls with center z

, = 1, . . . , N. The values
f(z

) : f T form a bounded set, so there exists m

such that [f(z

)[ m

for all f T. Let M


K
= 1 + max(m
1
, . . . , m
N
). Then if z K, there is an
such that [f(z)[ 1 +[f(z

)[ 1 +m

by taking [z

z[ < , for all f T. So


the sup above is bounded by M
K
.
Conversely, we want to show T is normal. We know T is normal if and
only if T is equicontinuous on each compact K and f(z) : f T is
bounded for all z . We know already that f(z) : f T is bounded for
each z by hypothesis, since z is compact for each z . So it remains to
show equicontinuity, which we will do using the Cauchy integral formula. Take
a compact K . For each z K, let
r
z
=
1
3
dist(z, ).
Take a nite subcover of K by balls
_
B(z
j
, r
zj
)
_
N
j=1
. Let r = min(r
1
, . . . , r
N
).
If z, w K and [z w[ < r, then there exists some z
j
such that [z z
j
[ < r
zj
,
128 Normal Families
so [z
j
w[ < 2r
zj
. Now
f(z) =
1
2i
_
|zj|=3rz
j
f() d
z
.
We have a similar formula for f(w), so
f(z) f(w) =
1
2i
_
|zj|=3rz
j
f()
_
1
z

1
w
_
d.
Hence
[f(z) f(w)[
1
2
2 3r
zj
[z w[
r
2
zj
max
|zj|=3rz
j
[f()[
=
3[z w[
r
zj
max
|zj|=3rz
j
[f()[.
Let
M
j
(f) = max
|zj|=3rz
j
[f()[ <
because f is continuous on compact sets. In fact,
M
j
:= sup
fF
M
j
(f) <
by hypothesis, and
M := max
j=1,...,N
M
j
< .
We conclude that
[f(z) f(w)[
3M
r
[z w[
for all f T. This gives equicontinuity on K. Given > 0, take < min(r,
r
3M
).
Then [z w[ < for all z, w K implies [f(z) f(w)[ < for all f T.
9.12 Remark. Notice that if T is normal, then f

: f T is also normal.
This is because the Cauchy integral formula says if f T is uniformly bounded
on compact sets. Use teh same circles as above adn that
[f

(z)[ =

1
2i
_
|zj|=3rz
j
f() d
( z)
2

3M
j
r
2
zj
Uniform Convergence 129
for z B(z
j
, 2r
zj
).
9.3 Riemann Mapping Theorem
We can apply Montels theorem to prove the Riemann mapping theorem.
9.13 Riemann Mapping Theorem. If C is a simply connected region
and ,= C, then for every z
0
there exists a unique bijective holomorphic
function f : D such that f(z
0
) = 0 and f(z
0
) > 0.
Proof. There are three steps. The rst of which is to prove the existence of
an analytic injection f : D such that z
0
0. In step two, we need to
prove that there exists an analytic injection f : D such that z
0
0 and
maximal [f

(z
0
)[. Lastly, we need to show that maximality of [f

(z
0
)[ implies f
is surjective.
Step 1. Take z
0
/ . Then z z
0
is analytic and nonzero on the simply
connected region , so there exists g(z) := log(z z
0
), an analytic branch of
the logarithm on , and e
g(z)
= z z
0
, which is injective on . We conclude
that g is injective. It also implies that g() omits quite a big piece of C. For
example, if we x w
0
and look at g(w
0
) + 2i, then the there is a constant
c > 0 such that g() w : [w (g(w
0
) + 2i)[ < c = .
1
Let
f(z) :=
1
g(z) (g(w) + 2i)
.
Then [f(z)[
1
c
, f is analytic, and f is injective because g is injective. Translate
so that f(z
0
) = 0 and divide by some large number so that f() D.
Step 2. Let T = f Hol(, D) : f injective, f(z
0
) = 0. This is a
nonempty set by Step 1. Notice that T is uniformly bounded by 1, so T is
normal. Let
s = sup
fF
[f

(z
0
)[.
Since f

(z
0
) ,= 0 for f from Step 1, we have s > 0. Take f
n
from T so
[f

n
(z
0
)[ s. Then T normal implies there exists a subsequence f
nm
con-
1
If not, then take {z
j
} such that g(z
j
) g(w
0
) + 2i. Then
e
g(z
j
)
= z
j
z
0
e
g(w
0
)+2i
= w
0
z
0
,
which implies z
j
w
0
. But g(z
j
) g(w
0
) by continuity, which contradicts that
g(z
j
) g(w
0
) + 2i.
130 Riemann Mapping Theorem
verging uniformly on compact sets in to f analytic on . Also, f is injective.
Otherwise there would be some z
1
,= z
2
so that f(z
1
) = f(z
2
). Then the se-
quence f
nm
(z) f
nm
(z
1
) is a sequence of analytic nonzero functions on z
1

converging uniformly on compact sets to f(z) f(z


1
), which has a zero on
z
1
. Hence f(z) f(z
1
) by Hurwitz theorem. Then f

(z
0
) = 0 and we
have a contradiction with [f

(z
0
)[ = s > 0. So f is injective, analytic and takes
z
0
0. Furthermore, [f

(z
0
)[ = s and [f
nm
(z)[ 1 for all z , so [f(z)[ 1
for all z and f is nonconstant. Hence f() D by the maximum principle.
Step 3. Suppose f() ,= D. Then there exists D f(). Let

(z) :=
z
1 z
.
Then (

f)() is simply connected and nonzero. So there exists a branch of


the square root on

f, call it h(z). Let


F :=
h((0))
h

f.
This is an injective analytic map D such that z
0
0. That is, F T
and s [F

(z
0
)[. Look at the relationship between f

(z
0
) and F

(z
0
). We have
f =
1

h
1

1
h((0))
F
and let :=
1

h
1

1
h((0))
. Then takes 0 0, D D, but it is not
injective. So the Schwarz lemma says [

(0)[ < 1. So f

(z
0
) =

(0)F

(z
0
) and
s = [f

(z
0
)[ < [F

(z
0
)[ s.
This is a contradiction, and were done.
CHAPTER 10
Selected Homework Solutions
10.1 Homework Set 1
1.1.1.2. If z = x +iy for x, y R, nd the real and imaginary parts of
z
4
,
1
z
,
z 1
z + 1
,
1
z
2
.
Solution:
Omitting details of the computation, with z = x +iy the answers are:
Re z
4
= x
4
6x
2
y
2
+y
4
, Imz
4
= 4xy(x
2
y
2
).
Re z
1
= x/(x
2
+y
2
), Imz
1
= y/(x
2
+y
2
).
Re(z 1)/(z + 1) = (x
2
+ y
2
1)/((x + 1)
2
+ y
2
), Im(z 1)/(z + 1) =
2y/((x + 1)
2
+y
2
).
Re z
2
= (x
2
y
2
)/(x
2
+y
2
)
2
, Imz
2
= 2xy/(x
2
+y
2
)
2
.
1.1.1.3. Show that
_
1 i

3
2
_
3
= 1 and
_
1 i

3
2
_
6
= 1
131
132 Homework Set 1
for all combinations of signs.
Solution:
This is an easy direct computation, or can be done by noting that all of these
numbers have modulus 1, the rst pair have argument

3
(so multiplication
by 3 gives multiples of 2), while the other pair in the second set have arguments

3
(so multiplication by 6 gives multiples of 2).
1.1.2.4. Solve the quadratic equation
z
2
+ ( +i)z + +i = 0.
Solution:
One may apply the quadratic formula.
1.1.4.3. Prove that

a b
1 ab

= 1
if either [a[ = 1 or [b[ = 1. What exception must be made if [a[ = [b[ = 1?
Solution:
If [a[ = 1 then

a b
1 ab

= [a[

a b
a [a[
2
b

= 1
and an analogous argument using b holds if [b[ = 1. The only exception is when
the denominator vanishes, which occurs i ab = 1. In the case that either [a[ = 1
or [b[ = 1 we conclude immediately that this occurs i a = b.
1.1.4.4. Find the conditions under which the equation az + bz + c = 0 in
one complex unknown has exactly one solution, and compute that solution.
Solution:
An equation in complex variables is two equations in real variables, in this
case two simultaneous linear equations. A fast way to extract them is to take
the complex conjugate, so we have
az +bz +c = 0
az +bz +c = 0.
Selected Homework Solutions 133
Multiplying by a and b respectively,
[a[
2
z +abz +ac = 0
abz +[b[
2
z +bc = 0
the dierence eliminates z and we obtain
_
[a[
2
[b[
2
_
z +
_
ac bc
_
= 0
so that there is a unique solution i [a[ , = [b[, in which case the solution is
z =
bc ac
[a[
2
[b[
2
1.1.5.1.Prove that

a b
1 ab

< 1
if [a[ < 1 and [b[ < 1.
Solution:
We have

a b
1 ab

2
< 1
a b
1 ab
a b
1 ab
< 1
[a[
2
+[b[
2
ba ab < 1 ab ab +[a[
2
[b[
2
[a[
2
_
1 [b[
2
_
< 1 [b[
2

_
[a[ < 1 and [b[ < 1
_
or
_
[a[ > 1 and [b[ > 1
_
.
1.1.5.3. If [a
i
[ < 1,
i
0 for 1 i n, and
1
+ +
n
= 1, show that
[
1
a
1
+ +
n
a
n
[ < 1.
Solution:
By induction and the triangle inequality

j=1

j
a
j

j=1

a
j

<
n

j=1

= 1.
134 Homework Set 1
1.1.5.4. Show that there are complex numbers z satisfying
[z a[ +[z +a[ = 2[c[
if and only if [a[ [c[. If this condition is fullled, what are the smallest and
largest values of [z[?
Solution:
It is easiest to solve this problem if you recognize it as the equation of an
ellipse with foci at a, as then you know what to expect. In any case it is easy
to see
2[a[ = [a +a[ = [a z +a +z[ [z a[ +[z +a[ = 2[c[
so that [a[ [c[ is necessary for the existence of z satisfying the equation.
Then for [c[ [a[ we may set z = [c[a/[a[ and nd that [z a[ = [c[ [a[,
[z +a[ = [c[ +[a[, so [z a[ +[z +a[ = 2[c[, whence it is also sucient.
To nd the maximum of [z[ on the curve is easy, because repetition of our
rst argument shows [z[ [c[ on the curve, but the point z = [c[a/[a[ has
[z[ = [c[, so this maximum is achieved. The minimum is a little trickier. One
way is as follows:
4[z[
2
= 4zz =
_
(za)+(z+a)
__
(za)+(z+a)
_
= [za[
2
+[z+a[
2
+2[z[
2
2[a[
2
so 2
_
[z[
2
[a[
2
_
= [z a[
2
+ [z + a[
2
. Then use the general inequality 2xy
x
2
+ y
2
for x, y R to see 2[z a[[z + a[ [z a[
2
+[z + a[
2
with equality i
[z a[ = [z +a[, and substitute into the square of the equation for 2[c[ to obtain
2[c[
2
[z a[
2
+ [z + a[
2
. Combining these we have 2[c[
2
2
_
[z[
2
[a[
2
_
, so
[z[
2
[c[
2
[a[
2
, with equality i [z a[ = [z +a[. The latter fact allows us to
guess that a point at distance
_
[c[
2
[a[
2
from 0 in a direction perpendicular
to a will lie on the curve, a guess that is rapidly veried using Pythagoras
theorem.
1.2.1.1. Find the symmetric points of a with respect to the lines which
bisect the angles between the coordinate axes.
Solution:
The easy way to do this is to note that symmetry is preserved under rotation
by /4, such that the lines become the axes, then use that the symmetric points
with respect to the axes are obtained by and conjugation, and rotate back.
Selected Homework Solutions 135
The points are then
e
i/4
(e
i/4
a) = a,
e
i/4
(e
i/4
a) = e
i/4
e
i/4
a = ia.
1.2.1.2. Prove that the points a
1
, a
2
, a
3
are the vertices of an equilateral
triangle if and only if a
2
1
+a
2
2
+a
2
3
= a
1
a
2
+a
2
a
3
+a
3
a
1
.
Solution:
The vertices of an equilateral triangle are mapped to the vertices of an
equilateral triangle by any map of the form z (z + ), and for any given
equilateral triangle there is a map of this form that takes it to the triangle
with vertices 1, e
2i/3
, e
2i/3
. The correct choice is 3 = a
1
+ a
2
+ a
3
and
= 3/(2a
1
a
2
a
3
). It is obvious that the equality
a
2
1
+a
2
2
+a
2
3
= a
1
a
2
+a
2
a
3
+a
3
a
1
is preserved by the map z z, and readily veried that it is preserved by
z z + . Thus it suces to verify the equality for the triangle with vertices
1, e
2i/3
, e
2i/3
, in which case both sides are easily checked to sum to 2.
1.2.2.2. Simplify 1 +cos +cos 2+ +cos n and sin +sin 2+ +
sin n.
Solution:
If z = e
i
then
1 + cos + + cos n = Re(1 +z + +z
n
)
= Re
1 z
n+1
1 z
= Re
1 z z
n+1
+z
n
1 +[z[
2
z z
=
1 cos cos(n + 1) + cos n
2 2 cos
sin + + sin n = Im
1 z z
n+1
+z
n
1 +[z[
2
z z
=
sin sin(n + 1) + sin n
2 2 cos
136 Homework Set 1
1.2.2.4. If
= cos
2
n
+i sin
2
n
,
prove that
1 +
h
+
2h
+ +
(n1)h
= 0
for any integer h which is not a multiple of n.
Solution:
Let w = cos(2i/n) +i sin(2i/n) = e
2i/n
. Provided h is not a multiple of
n we have w
nh
,= 1 and thus
1 +w
h
+. . . +w
(n1)h
=
1 w
nh
1 w
h
=
1 e
2hi
1 w
h
= 0.
1.2.3.1. When does az +bz +c = 0 represent a line?
Solution:
To see when az+bz+c = 0 represents a line, recall exercise 1.1.4.4. We must
have innitely many solutions for the simultaneous equations, so they must be
the same, so [a[ = [b[ and ac = bc. An equivalent form of the latter is that c
is a real multiple of (a + b); we may nd this by summing it with its complex
conjugate to nd (a +b)c = (a +b)c, from which the ratio (a +b)/c is real. To
be certain that the solution set is one dimensional not two dimensional we must
insist that [a[ , = 0.
1.2.3.5. Show that all circles which pass through a and 1/a intersect the
circle [z[ = 1 at right angles.
Solution:
Rotation around the origin preserves the circle [z[ = 1, preserves angles, and
may be used to move a to lie on the positive real axis. WLOG a > 1. Then the
circle through a and
1
a
has center
1
2
_
a +
1
a
_
and radius
1
2
_
a
1
a
_
. The circles
intersect at right angles i the radii to the intersection point are at right angles
i (by Pythagoras theorem) the sum of the squares of the radii of the circles
is the square of the distance to the center of the the circle through a and
1
a
.
Written algebraically, this is equivalent to
1 +
1
4
_
a
1
a
_
2
= 1 +
1
4
_
a
2
+
1
a
2
2
_
=
1
4
_
a
2
+
1
a
2
+ 2
_
=
1
4
_
a +
1
a
_
2
Selected Homework Solutions 137
1.2.4.5. Find the radius of the spherical image of the circle in the plane
whose center is a and radius R.
Solution:
I think it likely that this problem will have caused diculties. The reason
is that the spherical distance in equation (28) on page 20 of the book is not
a geodesic metric! It is the metric corresponding to chords of the sphere, so
the triangle inequality is always strict for three distinct points because they
cannot all lie on a chord (the chord intersects the sphere at two points only).
This and the fact that the center of the disc in the plane is not the same as the
center of the circle on the Riemann sphere mean that you cannot do either of the
following: (1) take the distance d(a, a+Ra/[a[), or d(a, aRa/[a[) (because a is
not the center in the spherical metric), (2) take half of d(aRa/[a[, a+Ra/[a[)
(because the metric is not geodesic, so half of the diameter is not the radius).
You can do one thing to simplify your calculations, which is to rotate the plane
(an isometry of the spherical metric) so a R. At this point you have the option
of nding the images on the sphere (using equations (25) and (26) on page 18),
then nd the center, then nd the length of the chord corresponding to the
radius. This latter can also be done by Pythagoras theorem, which makes the
problem much simpler.
Suppose that the length of the chord joining the images of aR on the sphere
is 2s. Then the distance from the center of the sphere to the midpoint of the
chord is

1 s
2
(by Pythagoras), so the distance radially from this midpoint to
the sphere is 1

1 s
2
. The landing point of the ray is the spherical center of
the disc, and the spherical radius we seek is the hypotenuse of the right triangle
with vertices the landing point, the midpoint of the chord, and the image of
a + R. Since we have just seen the sides of this triangle have lengths s and
1

1 s
2
the hypotenuse has length 2(1

1 s
2
) (Pythagoras again). So
it suces to nd s. By formula 28 on page 20,
2s =
2[(a +R) (a R)[
_
1 + (a +R)
2
__
1 + (a R)
2
_ =
4R
_
1 + (a +R)
2
__
1 + (a R)
2
_.
138 Homework Set 2
We can then compute
1 s
2
=
_
1 + (a +R)
2
__
1 + (a R)
2
_
4R
2
_
1 + (a +R)
2
__
1 + (a R)
2
_
=
1 + (a +R)
2
+ (a R)
2
+ (a
2
R
2
)
2
4R
2
_
1 + (a +R)
2
__
1 + (a R)
2
_
=
1 + 2(a
2
R
2
) + (a
2
R
2
)
2
_
1 + (a +R)
2
__
1 + (a R)
2
_
=
_
1 +a
2
R
2
_
2
_
1 + (a +R)
2
__
1 + (a R)
2
_
so that
Spherical radius of disc = 2(1
_
1 s
2
)
= 2
_
1

1 +a
2
R
2

_
1 + (a +R)
2
_
1/2
_
1 + (a R)
2
_
1/2
_
= 2
_
_
(1 +a
2
R
2
)
2
+ 4R
2
[1 +a
2
R
2
[
_
(1 +a
2
R
2
)
2
+ 4R
2
_
.
10.2 Homework Set 2
2.1.2.4. Show that an analytic function cannot have a constant absolute
value without reducing to a constant.
Solution:
Suppose f = u + iv is analytic and [f[ is constant, so [f[
2
= u
2
+ v
2
= c.
Analyticity gives existence of the partial derivatives u
x
, u
y
, v
x
, v
y
and validity
of the Cauchy-Riemann equations. Dierentiating [f[
2
= c gives uu
x
+ vv
x
=
uu
y
+vv
y
= 0; Cauchy-Riemann lets us rewrite these as
_
u v
v u
__
u
x
v
x
_
=
_
0
0
_
.
Since the determinant is [f[
2
= c, we discover that either c = 0, whence f = 0
everywhere, or the matrix is invertible and u
x
= v
x
= 0, from which also
u
y
= v
y
= 0 by Cauchy-Riemann and thus u and v, and hence f, are constant.
Selected Homework Solutions 139
2.1.2.5. Prove rigorously that the functions f(z) and f(z) are simultane-
ously analytic.
Solution:
Each of the following statements is equivalent.
f(z) is analytic at z = z
0
. lim
z0
f(z +z
0
) f(z
0
)
z
exists.
lim
z0
f(z +z
0
) f(z
0
)
z
exists.
lim
w0
f(w +w
0
) f(w
0
)
w
exists.
f(w) is analytic at w = w
0
,
where w = z and w
0
= z
0
.
2.1.4.1. Use the method of the text to develop
R(z) =
z
4
z
3
1
and r(z) =
1
z(z + 1)
2
(z + 2)
3
in partial fractions.
Solution:
Let R(z) = z
4
(z
3
1)
1
. It has poles at , 1, e
2i/3
, e
2i/3
. Set = 2i/3.
We need to expand at each pole, for which purpose (using the notation in the
book) perform each of the following expansions. We include only the relevant
terms, noting the presence of those with lower order using an ellipsis.
R(z) = z +
z
z
3
1
=G(z) = z
R(1 +z
1
) =
(1 +z
1
)
4
(1 +z
1
)
3
1
=
(z + 1)
4
z(z + 1)
3
z
4
=
z
4
+
3z
3
+
=
z
3
+
G
1
_
1
z 1
_
=
1
3(z 1)
140 Homework Set 2
R(e

+z
1
) =
(e

+z
1
)
4
(e

+z
1
)
3
1
=
(e

z + 1)
4
z(e

z + 1)
3
z
4
=
e
4
z
4
+
3e
2
z
3
+
=
e
2
z
3
+
G
e

_
1
z e

_
=
e
2
3(z e

)
R(e
2
+z
1
) =
(e
2
+z
1
)
4
(e
2
+z
1
)
3
1
=
(e
2
z + 1)
4
z(e
2
z + 1)
3
z
4
=
e
8
z
4
+
3e
4
z
3
+
=
e
4
z
3
+ =
e

z
3
+
G
e
2
_
1
z e
2
_
=
e

3(z e
2
)
Thus
R(z) = z +
1
3(z 1)
+
e
4i/3
3(z e
2i/3
)
+
e
2i/3
3(z e
4i/3
)
.
In a similar but more tedious calculation,
R(z) = z
1
(z + 1)
2
(z + 2)
3
has poles at 0, 1 and 2. We perform the following expansions, including only
the relevant terms.
R(z
1
) =
1
z
1
(z
1
+ 1)
2
(z
1
+ 2)
3
=
z
6
(1 +z)
2
(1 + 2z)
3
=
z
6
8z
5
+
=
z
8
+
G
0
_
1
z
_
=
1
8z
R(1 +z
1
) =
1
(1 +z
1
)z
2
(z
1
+ 1)
3
=
z
6
(z 1)(1 +z)
3
=
z
6
z
4
2z
3
+
= z
2
+ 2z +
G
1
_
1
(z + 1)
_
=
1
(z + 1)
2
+
2
(z + 1)
Selected Homework Solutions 141
R(2 +z
1
) =
1
(2 +z
1
)(1 +z
1
)
2
z
3
=
z
6
(2z + 1)(z + 1)
2
=
z
6
2z
3
+ 5z
2
4z +
=
z
3
2

5z
2
4

17z
8
+
G
2
_
1
(z + 2)
_
=
1
2(z + 2)
3

5
4(z + 2)
2

17
8(z + 2)
Thus
R(z) =
1
8z
+
1
(z + 1)
2
+
2
(z + 1)

1
2(z + 2)
3

5
4(z + 2)
2

17
8(z + 2)
.
2.1.4.4. What is the general form of a rational functions which has absolute
value 1 on the circle [z[ = 1? In particular, how are the zeros and poles related
to each other?
Solution:
Suppose R(z) is a rational function with [R(z)[ = 1 on [z[ = 1. Then
S(z) =
1
R
_
1
z
_
is also a rational function, and since w =
1
w
whenever [W[ = 1 we see that
R(z) = S(z) on [z[ = 1. But then R(z) = S(z), because their dierence is
a rational function with zeros at every point of the unit circle. Now observe
that if R(z) has a root at a of order then S(z) has a pole at z = (a)
1
with
the same order, so R(z) = S(z) has a pole of this order at this location. This
argument is reversible; if R(z) has a pole at b of order then S(z) has a zero
at z = (b)
1
with the same order, so R(z) = S(z) has a zero of this order at
this location. Thus the poles and zeros of R are bijectively paired by the map
z (z)
1
. (Note that this implies there are neither poles nor zeros at points of
[z[ = 1, because for these points z = (z)
1
and there cannot simultaneously be
a zero and a pole at a single point.) An equivalent formulation is that R(z) is
necessarily a product of factors of the form
za
1az
with [a[ , = 1. Since we proved
in Exercise 1.1.4.3 that such factors have modulus 1 on the unit circle, this
condition is also sucient. Grouping the factors and using negative powers
j
when a zero is outside the unit disc we see that the general form of the rational
142 Homework Set 2
function we seek is
R(z) = cz
k
m

j=1
_
z a
j
1 a
j
z
_
j
with k and
j
in Z and [a
j
[ < 1 for all j.
2.1.4.6. If R(z) is a rational function of order n, how large and how small
can the order of R

(z) be?
Solution:
Suppose R(z) = cP(z)/Q(z) is a rational function, where P and Q are monic
polynomials without common zeros. Let m = deg(P) and n = deg(Q). It will
be convenient at rst to assume neither P nor Q is constant. We have
R

(z) = c
P

(z)Q(z) P(z)Q

(z)
Q
2
(z)
.
It is an easy observation that deg(P

) = m 1 and deg(Q

) = n 1, thus
deg(P

Q) = deg(PQ

) m + n 1. Observe that the lead term in P

Q has
coecient n while that in PQ

has coecient m. If we assume m ,= n then


these terms cannot cancel, so deg(P

Q PQ

) m + n 1 with equality at
least when m ,= n. To nd the degree of R

(z) we must then cancel common


factors. Any root of Q with order is a root of P

Q with the same order, so to


be a root of P

QPQ

it must divide P or Q

. By assumption P and Q have


no common roots, so it must be a root of Q

. An easy computation shows that


Q and Q

have common roots if and only if Q has multiple roots, and that the
order of the root in Q

is 1. We conclude that Q

/Q is of the form S/T,


where T is the (monic) product of the distinct factors of Q and has order k n,
and deg S = k 1. It is then apparent that
R

(z) = c
P

(z)T(z) P(z)S(z)
Q(z)T(z)
with deg(P

T PS) m + k 1 (with equality if m ,= n), deg(QT) = n + k,


Selected Homework Solutions 143
and the numerator and denominator having no common roots. It follows that
deg(R

) maxm+k 1, n +k
= maxm1, n +k
maxm1, n +n
deg R +n
2 deg(R)
and that equality holds at the rst inequality if m ,= n, at the second if k = n
(i.e. all poles of R are distinct), at the third and fourth if deg R = n m. Thus
we have deg(R

) 2 deg(R) and conditions under which equality holds.


We may also get lower bounds. If m ,= n then we have equality in the
rst inequality above, so deg(R

) = maxm + k 1, n + k max m, n + 1
deg(R) because we assumed Q non-constant and thus k 1. If m = n then
deg(P

T PS) m+k1 < n+k = deg(QT), so deg(R

) = n+k deg(R)+1.
(Note this also implies that deg(R

) = 2 deg(R) when m = n = k
What remains are the special cases where either P or Q is constant. If Q
is constant and P is not then R is a polynomial, so deg(R

) = deg(R) 1.
If P is constant and Q is not then the above reasoning implies R

= S/QT,
so deg(R

) = deg(QT) = k + n > n = deg(R). If both are constant then


deg(R

) = .
We may summarize our results as follows. If R(z) is a non-constant rational
function, then deg(R)1 deg(R) 2 deg(R), with equality on the left i R is
a polynomial and equality on the right at least when the poles of R are distinct
and there is no pole at innity.
2.2.3.3. Show that the sum of an absolutely convergent series does not
change if the terms are rearranged.
Solution:
Suppose

j
[a
j
[ converges. A re-ordering of the sum is given by a bijection
: N N, where the new sum is

j
a
(j)
. Both sums converge absolutely;
set S =

j
a
j
and T =

j
a
(j)
. Given > 0 let N be so large that for any
K N, each of

j>K
[a
j
[ ,

jK
a
j

jK
a
(j)

.
144 Homework Set 2
Let M be so large that
_
(j) : j M
_
j N (Note that then M N.)
[S T[

jM
a
j

jM
a
(j)

+ 2

j>N
[a
j
[ + 2 3
where middle inequality used that the sum is over those j in
_
j M (j) : j M
_

_
(j) : j M j M
_
,
which does not include any j N.
2.2.3.4. Discuss completely the convergence and uniform convergence of the
sequence nz
n

n1
.
Solution:
For xed z with [z[ 1 we have [nz
n
[ = n[z[
n
, while if [z[ < 1 we have
n[z[
n
= exp(log n + nlog [z[) 0 because log [z[ < 0 and (nlog [z[)/ log n
as n (using, for example, LHopitals rule). Hence nz
n
converges
pointwise on [z[ < 1. A slight improvement is that on [z[ r < 1 we have
n[z[
n
nr
n
0, so that the convergence is uniform on this disc. Any compact
set that lies in [z[ < 1 is in such a disc, so we have nz
n
converges uniformly
to the zero function on any compact subset of the open unit disc. Moreover,
this is the largest collection of sets on which the convergence is uniform. To see
this, suppose S is a set on which nz
n
converges uniformly, from which we see
S is in the unit disc and n[z[
n
0. If S is not a compact subset of the disc
then it contains points z
k
such that [z
k
[ 1. If n is so large that n[z[
n
< 1/2
on S then the fact that n[z
k
[
n
n as k provides a contradiction.
2.2.3.6. If U = u
1
+ u
2
+ and V = v
1
+ v
2
+ are convergent series,
prove that
UV = u
1
v
1
+ (u
1
v
2
+u
2
v
1
) + (u
1
v
3
+u
2
v
2
+u
3
v
1
) +
provided that at least one of the series is absolutely convergent. (It is easy if
both series are absolutely convergent. Try to arrange the proof so economically
that the absolute convergence of the second series is not needed.)
Solution:
There are a number of ways of doing this, the most usual being to use
Selected Homework Solutions 145
summation by parts. I will show a method that is messier, but perhaps easier
to visualize. Let U =

j
u
j
, V =

j
v
j
; suppose WLOG that

j
u
j
is the
absolutely convergent series and set W =

j
[u
j
[. It is helpful to arrange the
terms in a doubly-innite list.
(u
j
v
k
)
j,k
=
_
_
_
_
_
_
u
1
v
1
u
1
v
2
u
1
v
3
. . .
u
2
v
1
u
2
v
2
u
2
v
3
. . .
u
3
v
1
u
3
v
2
u
3
v
3
. . .
.
.
.
.
.
.
.
.
.
.
.
.
_
_
_
_
_
_
We see immediately that

n1
j=1
u
j
v
nj
is the sum along the n
th
upward diagonal,
so
m

n=1
n1

j=1
u
j
v
nj
is the sum over the upper left triangle j + k m in the list. We can also
see that UV may be approximated by the sum over a square j p, k p,
because for xed > 0 we can take N so p N implies

jp
u
j

kp
v
k

, and so

UV
_

jp
u
j
__

kp
v
k
_

UV V

jp
u
j
+V

jp
u
j

_

jp
u
j
__

kp
v
k
_

[V [ +

jp
u
j


_
[V [ +

jp
[u
j
[
_

_
[V [ +W
_
.
Thus the dierence between UV and the sum we are considering is controlled
by a small term plus the sum over a region inside an upper left triangle and
outside a square, i.e. j +k m but also j p, k p.

UV
m

n=1
n1

j=1
u
j
v
nj


_
[V [ +W
_
+

n=1
n1

j=1
u
j
v
nj

_

jp
u
j
__

kp
v
k
_


_
[V [ +W
_
+

{jp,kp,j+km}
u
j
v
k

.
146 Homework Set 2
We split this sum into two pieces, one with k p and one with k p. We
want a bound independent of m, and you should think that m p. Since it is
a nite sum it can be rearranged any way we like; we will sum rst along the
rows and then down the columns. For the piece with k p the sum along the
j
th
row is bounded as follows:

u
j
mj

k=p+1
v
k

[u
j
[
provided p is so large that

p+1
v
k


for all q p; this can be achieved by (if necessary) increasing N, because

v
k
is convergent. Summing over all relevant rows we have a contribution bounded
as follows:

mp

j=1
u
j
mj

k=p+1
v
k


mp

j=1
[u
j
[

j=1
[u
j
[ = W.
Now for the piece with k < p we reason as follows. Since

k
v
k
converges,

kr
v
k
is a convergent sequence in r, so has absolute value bounded by a
constant X. Now the sum across each row in the second piece is of size

u
j
min{p1,mj}

k=1
v
k

X[u
j
[
and summing over the rows we have

j=p+1
u
j
min{p1,mj}

1
v
k

X
m

j=p+1
[u
j
[ X

j=p+1
[u
j
[ X
provided p is large enough that

jp+1
[u
j
[ ;
again this can be achieved by increasing N. Combining our estimates we now
have

UV
m

n=1
n1

j=1
u
j
v
nj


_
[V [ + 2W +X
_
Selected Homework Solutions 147
and the result follows.
By the way, if you are wondering how this is connected to the material in
this chapter, the following may be of interest. Consider the functions
U(z) =

j
u
j
z
j
, V (z) =

j
v
j
z
j
.
These converge at z = 1, so have radius of convergence at least 1; moreover
Abels limit theorem says that U(r) U(1) and V (r) V (1) as r 1,
where r (0, 1). On the disc [z[ < 1 they are absolutely convergent, and their
product U(z)V (z) has a power series expansion

j
c
n
z
n
. We can compute by
the Leibniz rule
c
n
=
1
n!
d
n
dz
n
_
U(z)V (z)
_

z=0
=
1
n!
n

j=0
n!
j!(n j)!
d
j
U(z)
dz
j

z=0
d
nj
V (z)
dz
nj
=
n1

j=1
u
j
v
nj
.
If we now knew that

n
c
n
=

n
n1

j=1
u
j
v
nj
was convergent, then Abels limit theorem would imply

n
c
n
= lim
r1
U(r)V (r) = lim
r1
U(r) lim
r1
V (r) = U(1)V (1),
which is the theorem we seek. This suggests that there is a proof essentially by
the proof of Abels limit theorem, which is the standard summation by parts
proof.
2.2.4.2. Expand
2z + 3
z + 1
in powers of z 1. What is the radius of conver-
gence?
Solution:
148 Homework Set 2
2z + 3
z + 1
=
2(z 1) + 5
(z 1) + 2
= 2 +
1
2
1
1 + (z 1)/2
= 2 +
1
2

j=0
(1)
j
2
j
(z 1)
j
if [z 1[ < 2.
The radius of convergence is 2.
2.2.4.4. If

n0
a
n
z
n
has radius of convergence R, what is the radius of
convergence of

n0
a
n
z
2n
and

n0
a
2
n
z
n
?
Solution:
The following statements are equivalent.

n
a
n
z
n
has radius of convergence R
limsup [a
n
[
1/n
= R
1
limsup [a
n
[
1/2n
= R
1/2
and limsup [a
n
[
2/n
= R
2

n
a
n
z
2n
has radius of convergence

R, and

n
a
2
n
z
n
has radius of convergence R
2
2.2.4.8. For what values of z is

n0
_
z
1 +z
_
n
convergent?
Solution:
The series

n=0
z
n
(1+z)
n
converges if and only if [z(1+z)
1
[ < 1, which
is if and only if [z[ < [1 +z[, if and only if Re(z) >
1
2
.
2.3.2.2. The hyperbolic cosine and sine are dened by
cosh z =
e
z
+e
z
2
, sinh z =
e
z
e
z
2
.
Express them through cos iz and sin iz. Derive the addition formulas, and for-
Selected Homework Solutions 149
mulas for cosh 2z and sinh 2z.
Solution:
cos iz =
e
z
+e
z
2
= cosh z
sin iz =
e
z
e
z
2i
= i sinh z
We can get addition formulae for cosh and sinh by computing
cosh a cosh b =
e
a
+e
a
2
e
b
+e
b
4
=
e
a+b
+e
(a+b)
+e
ab
+e
ba
4
sinh a sinh b =
e
a
e
a
2
e
b
e
b
4
=
e
a+b
+e
(a+b)
e
ab
e
ba
4
cosh a sinh b =
e
a
+e
a
2
e
b
e
b
4
=
e
a+b
e
(a+b)
e
ab
+e
ba
4
elementary algebra implies
cosh(a +b) = cosh a cosh b + sinh a sinh b cosh 2a = cosh
2
a sinh
2
a
sinh(a +b) = sinh a cosh b + cosh a sinh b sinh 2a = 2 sinh a cosh a
and substitution of the above expressions for cos and sin retrieves the usual
trigonometric addition formulae, which could alternatively be used to obtain
the above.
2.3.2.3. Use the addition formulas to separate cos(x + iy) and sin(x + iy)
into real and imaginary parts.
Solution:
cos(x +iy) = cos xcos iy sin xsin iy = cos xcosh y i sin xsinh y
sin(x +iy) = sin xcos iy + cos xsin iy = sin xcosh y +i cos xsinh y
2.3.4.4. For what values of z is e
z
equal to 2, 1, i, i/2, 1 i, 1 + 2i?
Solution:
150 Homework Set 2
e
z
= 2 when z = log 2 + 2ki, k Z
e
z
= 1 when z = (1 + 2k)i, k Z
e
z
= i when z =
_
1
2
+ 2k
_
i, k Z
e
z
=
i
2
when z = log 2 +
_
3
2
+ 2k
_
i, k Z
e
z
= 1 i when z =
1
2
log 2 +
_
5
4
+ 2k
_
i, k Z
e
z
= 1 + 2i when z =
1
2
log 5 +
_
arctan 2 + 2k
_
i, k Z and arctan in (0, /2).
2.3.4.5. Find the real and imaginary parts of exp(e
z
).
Solution:
Let z = x + iy, so e
z
= e
x
cos y + ie
x
sin y. Then the real and imaginary
parts of exp(e
z
) are obtained from
exp(e
z
) = exp(e
x
cos y +ie
x
sin y)
= exp(e
x
cos y) cos(e
x
sin y) +i exp(e
x
cos y) sin(e
x
sin y)
2.3.4.6. Determine all the values of 2
i
, i
i
, and (1)
2i
.
Solution:
2
i
= exp(i log 2) = cos(log 2) +i sin(log 2) single valued because 2 R
i
i
= exp(i log i) = exp
_

2
2k), k Z
(1)
2i
= exp(2i log(1)) = exp(2 4k), k Z.
Note that the last example shows something we have lost in making the con-
vention that the logarithm is single valued on the positive reals and multivalued
elsewhere, because (1)
2i
,= ((1)
2
)
i
= 1, but simply contains 1 as one of its
values.
Selected Homework Solutions 151
10.3 Homework Set 3
3.1.2.6. A set is said to be discrete if all its points are isolated. Show that
a discrete set in R or C is countable.
Solution:
R and C are separable metric spaces, so see Exercise 3.1.3.7 below.
3.1.3.4. Let A be the set of points (x, y) R
2
with x = 0 and [y[ 1, and
let B be the set with x > 0 and y = sin(1/x). Is A B connected?
Solution:
Let A = (0, y) : [y[ 1 and B = (x, sin(1/x) : x > 0 in R
2
. Note
that both A and B are continuous images of connected subsets of R, so are
connected. If there is a disconnection of A B, then the set containing a point
of A must contain all of A because A is connected, and similarly for B. So the
disconnection must consist of open sets U A and V B. But then U contains
a disc around 0, and any such disc contains points from B, for example those of
the form
_
(2n)
1
, sin(2n)
_
for all suciently large n N, so U V ,= and
there is no such disconnection. We conclude that A B is connected. (Note:
A B is not path connected).
3.1.3.5. Let E be the set of points (x, y) R
2
such that x [0, 1] and
either y = 0 or y = 1/n for some positive integer n. What are the components
of E? Are they all closed? Are they relatively open? Verify that E is not locally
connected.
Solution:
Let E
n
= (x, 1/n) : 0 x 1 and E

= (x, 0) : 0 x 1 in R
2
. Let
E = E

n
E
n
_
. For any n let U
n
be the open set consisting of all points
within distance (n + 1)
2
of E
n
, and V
n
be the interior of the complement of
U
n
. Then U
n
V
n
= , U
n
E
n
and V
n
E E
n
. We conclude that the
maximal connected set containing a point of E
n
is contained in E
n
. Since E
n
is connected (because it is a curve), we conclude that E
n
is a component of E
for each n. For a point in E

we see that the component containing this point


must contain E

, because E

is connected; from the above it does not contain


any point of E
n
for any n, so E

is a component. We have therefore found the


decomposition of E into its components. We have also seen that the components
152 Homework Set 3
of the form E
n
are closed in R
2
and relatively open in E; the component E

is closed in R
2
, but not relatively open in E because any open neighborhood of
E

in R
2
intersects some E
n
.
Finally, a set is locally connected if every neighborhood of a point in the
set contains a connected neighborhood of the point. Consider (0, 0) E

.
Any neighborhood of this point is the intersection of E with an open set in
R
2
, and therefore cannot be connected because it contains points from another
component E
n
. Since this point has no connected neighborhoods it is not locally
connected. (Remark: local connectivity of a set is equivalent to all components
being open.)
3.1.3.7. A set is said to be discrete if all its points are isolated. Show that
a discrete set in a separable metric space is countable.
Solution:
S is discrete in a metric space (M, d), so for all x S there is r
x
such
that the ball around x of radius r
x
, denoted B(x, r
x
), has B(x, r
x
) S = x.
Observe that then B(x, r
x
/2) B(y, r
y
/2) = if x ,= y, because if there is z in
the intersection we would have
[x y[ [x z[ +[y z[ < (r
x
+r
y
)/2 maxr
x
, r
y

contradicting either x B(y, r


y
) or y B(x, r
x
).
Now if A M is a dense set then for each x there is a
x
B(x, r
x
/2) A,
and a
x
,= a
y
if x ,= y because B(x, r
x
/2) B(y, r
y
/2) = . The map x a
x
is
then an injection from S to A, so the cardinality of S cannot exceed that of A.
In particular if M is separable we may take A to be a countable dense subset
and conclude S is countable.
3.2.2.1. Give a precise denition of a single-valued branch of

1 +z +

1 z in a suitable region, and prove that it is analytic.


Solution:
We want to give a domain and a denition of

so that

1 +z +

1 z
is a single-valued analytic function. In the book there is a denition of a single
valued analytic branch of

w for w C(, 0]. It suces then that we ensure


1 + z and 1 z are in this set, which is true if z C
_
(, 1] [1, )
_
.
Nothing more need be done, as on this set

1 +z +

1 z is a sum of single-
Selected Homework Solutions 153
valued analytic functions.
10.4 Homework Set 4
3.3.1.1. Prove that the reection z z is not a linear transformation.
Solution:
The map z z xes 0,1 and . Consider a linear fractional transformation
z
az +b
cz +d
xing these points. Fixing 0 implies b = 0, xing implies c = 0, xing
1 implies a/d = 1, so the map is the identity. Thus z z is not a linear
fractional transformation.
3.3.3.1. Prove that every reection carries circles into circles.
Solution:
A reection z z

= g(z) is dened via


(z

, z
1
, z
2
, z
3
) = (z, z
1
, z
2
, z
3
).
Let be the linear fractional transformation so (z
1
) = 1, (z
2
) = 0 and
(z
3
) = , and let f(z) = z. Then (z

) = (z), so g =
1
f . Now
we know from the book that maps circles to circles (on the Riemann sphere),
so it suces to see that the same is true of f. A circle not through has the
form [z a[ = r, so is mapped to r = [z a[ = [z a[, which is also a circle. A
circle through is z = x +iy with ax +by +c = 0, a, b, c R, and is mapped
to z = x +iy with ax by +c = 0, which is also a circle through .
3.3.3.5. Find the most general linear transformation of the circle [z[ = R
into itself.
Solution:
It is easy to think of some linear fractional transformations that x [z[ = R
as a set. For example, any rotation around the origin (i.e. z az with [a[ = 1),
154 Homework Set 4
the map z R
2
/z. In the case R = 1 we also know from prior homeworks that
z
z a
1 az
for [a[ , = 1 preserves [z[ = 1, so
z R
(z/R) a
1 a(z/R)
=
Rz aR
2
R az
preserves [z[ = R. The more dicult question is whether we have found all
of the maps or not. One way to approach the problem is to take an arbitrary
linear fractional transformation that xes [z[ = R, compose it with some of the
maps we know, and try to get the result to x 3 points, so that the composition
is the identity.
Suppose that is a linear fractional transformation that xes [z[ = R as
a set. Let e
i
= (R)/R, so z e
i
(z) xes the point z = R. Let a =
e
i
(0)/R and
(z) =
Rz aR
2
R az
.
Then z (e
i
(z)) xes 0 and R, and also xes [z[ = R as a set. Since the
mapping of reection in a circle depends only on the circle, this map must pre-
serve pairs of points that are symmetric under reection in [z[ = R. Now
0, has such reective symmetry and (e
i
(0)) = 0, so we conclude
(e
i
()) = . Thus we have a linear fractional transformation xing
three points, which we know is the identity, and so (e
i
(z)) = z, from
which (z) = e
i

1
(z). It is easy to compute that

1
(z) =
Rz +aR
2
R +az
.
We have thus shown that
(z) = e
i
Rz +aR
2
R +az
.
3.3.3.7. Find a linear transformation which carries [z[ = 1 and [z
1
4
[ =
1
4
into concentric circles. What is the ratio of the radii?
Solution:
If we had a linear fractional transformation taking [z[ = 1 and [z
1
4
[ =
Selected Homework Solutions 155
1
4
to concentric circles, then by composing with a translation to move their
common center to 0 and a dilation we could assume that the linear fractional
transformation preserves the unit circle; note that neither operation aects the
ratio of radii of the image circles. By Exercise 3.3.3.5 with R = 1, such a map
is of the form
z e
i
z a
1 az
= f(z), [a[ , = 1.
The rotation preserves both the concentric property and the ratio of radii, so
we are free to choose it, for example by setting = 0. Now observe that the
initial and nal congurations are mapped to themselves by z z, so
g(z) = f(z) = e
i
z a
1 az
also preserves [z[ = 1 and takes [z
1
4
[ =
1
4
to a circle with center at 0. If the
image of [z
1
4
[ =
1
4
under f is [z[ = r, it follows from Exercise 3.3.3.5 that
f g
1
(z) =
z+a
1+az
a
1 a
z+a
1+az
=
(1 a
2
)z + (a a)
(1 [a[
2
) + (a a)z
=
1 a
2
1 [a[
2
z +
aa
1a
2
1 +
aa
1|a|
2
z
is a linear fractional transformation xing [z[ = r, so is of the form
z e
i
rz +r
2
b
r +bz
.
We conclude that a R. Knowing this, we see that f preserves R, and therefore
that f(0) and f(
1
2
) are the points r. This gives us that simultaneously a =
r and (
1
2
a)/(1
a
2
) = r, which we reduce to r
2
4r+1 = 0, or (nding only
solutions with r > 0), a = r = 2

3. We have found necessary conditions for f


to map [z
1
4
[ =
1
4
to [z[ = r, and it remains to see they are sucient. But f is
a linear fractional transformation so maps circles to circles; by construction the
image of [z
1
4
[ =
1
4
passes through r, and since a R the image is mapped
to itself by z z, so it is [z[ = r. We have therefore determined that the map
is one of
z e
i
z a
1 az
for a = 2

3 and the ratio of the smaller to larger radius is 2

3 : 1 = 1 :
2 +

3.
3.3.3.8. Find a linear transformation which carries [z[ = 1 and Re(z) = 2
156 Homework Set 4
into concentric circles. What is the ratio of the radii?
Solution:
We are now asked to repeat this problem for [z[ = 1 and x = 2. All of the
same arguments apply to say that the map is
z e
i
z a
1 az
for a R. The points z = 2 and z = are invariant under z z, so go to
r. Computing as in the previous problem we nd again r = 2

3, but with
a =
1
r
= 2

3. Alternatively we could notice that the map z


1
z
takes the
circle in 3.3.3.7 to the line in this problem (to see this, note what happens to 0
and and that the symmetry under z z is preserved), so the whole problem
is obtained from 3.3.3.7 by composition with z
1
z
.
3.4.2.2. Map the region between [z[ = 1 and [z
1
2
[ =
1
2
on a half plane.
Solution:
The circles intersect at z = 1 and are parallel there, so any linear fractional
transformation mapping 1 to will give us the region between two circles that
intersect at with zero angle, meaning a strip between two parallel lines. There
are many linear fractional transformations taking 1 to , but it is convenient
to take one that maps one of the circles to the real axis; for example
z i
z + 1
z 1
takes [z[ = 1 to R because it takes 1 to 0 and i to 1. The image of the other
line must have the form z = x + ic for a constant c; by computing 0 i we
nd c = i. Now we can multiply by and exponentiate to get the upper
half plane. Our nal map is
z exp
_
i
z + 1
z 1
_
.
3.4.2.3. Map the complement of the arc [z[ = 1, Im(z) 0 on the outside
of the unit circle so the that points at correspond to each other.
Solution:
The arc [z[ = 1, y 0 can be mapped to a segment on the real line by a
Selected Homework Solutions 157
linear fractional transformation, and we know how to map the complement of
a segment on the complement of the unit disc (or rather we know how to map
the complement of the unit disc on the complement of the segment [2, 2] using
the map z z + z
1
). For the rst step we should take a point on [z[ = 1 to
, and since it is convenient to make the image segment symmetric around 0
we may as well take i to 0 and i to . After multiplying by 2 so our arc goes
to [2, 2], the map is z 2i
zi
z+i
.
Now the trickier issue is how to invert z z + z
1
. Formally using the
quadratic equation the inverse is z
1
2
(z

z
2
4), but we have to make
sense of the square root. If z
2
4 mapped the complement of [2, 2] to the
complement of a ray connecting 0 and we could use a version of the usual
square root, but it does not the image in this case is the complement of the
interval [4, 0]. However, we can do whatever algebraic manipulation we desire
to the expression for the formal inverse and still have a formal inverse. Knowing
that we want the bit inside the square root to omit a ray from 0 to (preferably
the negative real axis) and that right now what is in there omits [4, 0], tends
to suggest we should divide by either z or z 2 to move an endpoint to . A
little playing with the formula yields a dierent formal expression for the inverse
branches
f

(z) =
z
2
_
1
_
1 4/z
2
_
in which 14/z
2
is easily seen to omit the negative real axis (and the ray (2, )
on the positive real axis). We can therefore use the usual denition of

taking
C (, 0] to the right half plane. The result is that the branches f

(z) are
each well dened on the complement of [2, 2].
Now note that the two branches of the inverse map either to the interior
or the exterior of the unit disc, and that we want the one which maps to the
exterior. We can verify that the desired one is
z
2
(z +

1 4z
2
) by observing
that it has image converging to as z .
Our nal result is therefore that
z
1
2
_
2i
z i
z +i
_
_
1 +
_
1
4(z +i)
2
4(z i)
2
_
1/2
_
=
1 +iz
z +i
_
1 +
_
1 +
(z +i)
2
(z i)
2
_
1/2
_
where we readily determine that for z not on our arc, 1 +
(z+i)
2
(zi)
2
is not on the
negative real axis.
3.4.2.7. Map the outside of the ellipse (
x
a
)
2
+ (
y
b
)
2
= 1 onto [w[ < 1 with
158 Homework Set 4
preservation of symmetries.
Solution:
We wish to map the exterior of an ellipse on the interior of the unit disc
with preservation of symmetries, those being z z and z z. Recall that in
studying the map z z+z
1
= f(z) we saw that the circle z = re
i
, is mapped
to z = x+iy with x = (r +r
1
) cos and y = (r r
1
) sin , which is an ellipse
centered at 0 with semi-major axis along the real line and of length a = r +r
1
and semi-minor axis along the imaginary axis and of length b = [r r
1
[; this is
the ellipse (x/a)
2
+ (y/b)
2
= 1, a > b. The map also preserves the symmetries
of the ellipse, in that f(z) = f(z) and f(z) = f(z).
The map z f(z) is a double covering; if r > 1 then both [z[ > r and
[z[ < 1/r are mapped onto the the exterior of the ellipse. It is easy to see that
when r > 1 we should have r = (a b)/2 and thus 1/r = 2/(a b). The
inverse map has two branches z
z
2
(1

1 4z
2
) which were discussed in
the previous problem. Here we need the branch z
z
2
(1

1 4z
2
) which
maps the exterior of the ellipse to the interior of the disc of radius 2/(a b).
There are only a few more diculties. The rst is that we need the ellipse to
have a > b, so that a preliminary rotation is needed if b > a. The second is
that we need r = (a b)/2 > 1 in the above analysis. To ensure that this is the
case it is convenient to perform a preliminary dilation, for example one ensuring
that r = 2. Then the inverse branch of f will map to the disc [z[ < 1/2, and an
additional dilation will bring us to [z[ < 1. At this point it is simplest to split
into cases.
Case 1: a > b. Then (x/a)
2
+ (y/b)
2
= 1 has its longer axis along the real
line. We perform the dilation
z
4z
a b
,
so that the length of the semi-major axis becomes 4a/(a b) and of the semi-
minor axis becomes 4b/(a b); the exterior of this is the image of [z[ < 1/2
under f(z), so applying
z
z
2
(1
_
1 4z
2
)
takes it to [z[ < 1/2 and dilating by a factor of 2 gives [z[ < 1. The result is
z 2
4z
2(a b)
_
1
_
1
4(a b)
2
16z
2
_
1/2
_
=
4z
a b
_
1
_
1
(a b)
2
4z
2
_
1/2
_
Selected Homework Solutions 159
Case 2: b > a. We can use the same argument as above, but must rst
rotate by /2 to get the semi-major axis along the real line, and must undo this
at the end. The maps are
z iz
4iz
b a
,
then the inverse branch, then dilation by 2 and rotation by i, resulting in
z 2i
4iz
2(b a)
_
1
_
1
4(b a)
2
16z
2
_
1/2
_
=
4z
b a
_
1
_
1 +
(a b)
2
4z
2
_
1/2
_
Case 3: b = a. Here we have just the circle [z[ = a = b, so we can scale by
z z/a to the unit circle and invert to map the exterior to the interior. The
resulting map is
z
a
z
.
10.5 Homework Set 5
4.1.3.2. Compute
_
|z|=r
x dz =
_
|z|=r
Re(z) dz,
for the positive sense of the circle, in two ways: rst, by use of a parameter,
and second, by observing that x =
1
2
(z +z) =
1
2
(z +
r
2
z
) on the circle.
Solution:
Using the parametrization, z = re
i
, so z = x+iy with x = r cos , y = r sin
we have
_
|z|=r
x dz =
_
2
0
r cos ire
i
d
= ir
2
_
2
0
cos
2
+i cos sin d
= ir
2
_
2
0
1
2
(cos 2 + 1) d r
2
_
2
0
1
2
sin 2 d
= ir
2
.
Using
x =
1
2
(z +r
2
z
1
)
160 Homework Set 5
as suggested in the text we use that
_
|z|=r
z dz = 0
by equation (11) on pg 107 and
_
|z|=r
1
z
dz = 2i
by the following discussion to conclude
_
|z|=r
x dz = ir
2
.
4.1.3.3. Compute
_
|z|=2
dz
z
2
1
for the positive sense of the circle.
Solution:
The function (z
2
1)
1
does not have an antiderivative on the disc [z[ 2,
so we cannot directly apply the results from section 3.3. Since we do not yet
have the Cauchy theorem, we must integrate directly using a parameter. This is
messy, but the point is to get you to appreciate the Cauchy theorem and residue
calculus once we have them. Take z = 2e
i
, so the integral is
_

2ie
i
4e
2i
1
d =
_

2ie
i
(4e
2i
1)
(4e
2i
1)(4e
2i
1)
d
=
_

8ie
i
2ie
i
17 8 cos 2
d
=
_

6i cos + 10 sin
17 8 cos 2
d.
The real part of the integral has an antisymmetric integrand on a symmetric in-
terval so is zero. For the imaginary part we note from symmetry and periodicity
Selected Homework Solutions 161
of cosine that
_

cos
17 8 cos 2
d =
_

0
2 cos
17 8 cos 2
d
=
_
/2
0
2 cos
17 8 cos 2
d +
_

/2
2 cos
17 8 cos 2
d
=
_
/2
0
2 cos
17 8 cos 2
d +
_
0
pi/2
2 cos( )
17 8 cos(2 2)
(d)
=
_
/2
0
2 cos
17 8 cos 2
d +
_
/2
0
2 cos()
17 8 cos(2)
d
= 0
so that the whole integral is zero.
4.1.3.6. Assume that f(z) is analytic and satises the inequality [f(z)1[ <
1 in a region . Show that
_

(z)
f(z)
dz = 0
for every closed curve in . (The continuity of f

(z) is taken for granted.)


Solution:
We know that f

/f should have antiderivative log f, provided that log f is


well dened. Suppose then that [f(z) 1[ < 1 for z . We know that log is a
single-valued analytic function on C (, 0], with derivative z
1
. Since the
values of f avoid (, 0] it follows that log f is well dened and analytic, with
derivative (by the chain rule) f

(z)/f(z). Hence the integrand is the derivative


of an analytic function on and by the result on page 107 the integral around
any closed curve in is zero.
4.1.3.8. Describe a set of circumstances under which the formula
_

log(z) dz = 0
is meaningful and true.
Solution:
In order that
_

log z dz = 0 is meaningful and true, we must rst require


that lies in a region where log z is a single-valued function (otherwise the
integral is not well-dened). We know = C(, 0] is such a region (in fact,
162 Homework Set 6
so is the complement of any connected set containing 0 and ). Now we might
hope to nd that
_

log z dz = 0 for any closed curve in by proving that log z


is the derivative of an analytic function. Since we know that (log z)

= z
1
on
, it is easy to check that (z log z z)

= log z +z/z 1 = log z there. Hence a


suitable condition is that is a closed curve in .
4.2.2.1. Compute
_
|z|=1
e
z
z
dz.
Solution:
Using Cauchys integral formula we have
_
|z|=1
e
z
z
dz = 2ie
0
= 2i.
4.2.2.2. Compute
_
|z|=2
dz
z
2
+ 1
using partial fractions decomposition.
Solution:
Writing
1
z
2
+ 1
=
1
(z +i)(z i)
=
1
2i
_
1
z i

1
z +i
_
we can apply the Cauchy integral formula to each of the below integrals
_
|z|=2
1
z
2
+ 1
dz =
1
2i
_
|z|=2
1
z i
dz
1
2i
_
|z|=2
1
z +i
dz = = 0.
Compare this to exercise 4.1.3.3 above!
10.6 Homework Set 6
4.2.3.1. Compute
_
|z|=1
e
z
z
n
dz,
_
|z|=2
z
n
(1 z)
m
dz,
_
|z|=
[z a[
4
[dz[, ([a[ , = ).
Solution:
Selected Homework Solutions 163
(a) By the Cauchy formula for derivatives,
_
|z|=1
z
n
f(z) dz =
2if
(n1)
(z)
(n 1)!
provided f is analytic in the unit disc. Applying this to f(z) = e
z
gives
_
|z|=1
z
n
e
z
dz =
2i
(n 1)!
.
(b) If n and m are both non-negative then the integrand in
_
|z|=2
z
n
(1 z)
m
dz
is a polynomial, hence analytic, and so the integral is zero by the Cauchy for-
mula. If n is negative and m non-negative then our previous reasoning says the
result is 2if
(|n|1)
(0)/([n[1)! for f(z) = (1z)
m
, which is 0 if m < [n[1 and
2i(1)
|n|1
_
m
|n|1
_
otherwise. Similarly, if n is non-negative and m negative
then we get 0 if n < [m[ 1 and
2ig
(|m|1)
(1)
([m[ 1)!
= 2i
_
n
[m[ 1
_
otherwise (with g(z) = z
n
). If both m and n are negative it is a little more
work, basically because we do not have the Cauchy formula for general curves,
so cannot reduce to a circle around 0 and another around 1. However there is a
still a relatively quick argument of the same type as above. We may use partial
fractions to write z
n
(1 z)
m
= P(z)z
n
+Q(z)(1 z)
m
, where P has degree at
most [n[ 1 and Q has degree at most [m[ 1; note that
P(z)(1 z)
m
+Q(z)z
n
= 1.
Integrating the P(z)z
n
term gives 2iP
(|n|1)
(0)/([n[ 1)!, which is just 2i
times the leading coecient p of P(z). Similarly integrating the Q(z)(1 z)
m
term gives (1)
m
2i times the leading coecient q of Q. These leading terms
give the power of z
mn1
in P(z)(1 z)
m
+ Q(z)z
n
to be (1)
m
(p + q),
and this must be zero because m, n < 0 implies mn 2, so mn1 1.
164 Homework Set 6
We may summarize our results as
_
|z|=2
z
n
(1 z)
m
dz =
_

_
2i(1)
n1
_
m
n1
_
if n < 0 and m n 1
2i
_
n
m1
_
if m < 0 and n m1
0 otherwise.
(c) We write [z a[
2
= (z a)(z a) = (z a)
_
(
2
/z) a
_
on [z[ = and
use the fact that on z = e
i
we have [dz[ = d and dz = ire
i
d = iz d, so
[dz[ = (dz/iz). Thus the integral is
_
|z|=
[z a[
4
[dz[ =
_
|z|=
1
(z a)
2
_

2
z
a
_
2

iz
dz
=

i
_
|z|=
z
(z a)
2
(
2
az)
2
dz.
Now by assumption, [a[ , = . The integrand has singularities at z = a and
z =
2
/a, which are reection-symmetric in the circle [z[ = , so only one is
inside the circle. The Cauchy formula then says that the integral is 2f

(a)
with f(z) = z/(
2
az)
2
if [a[ < , which evaluates to 2(
2
+[a[
2
)/(
2
[a[
2
)
3
.
Similarly if [a[ > it becomes 2g

(
2
/a) with g(z) = z/(a)
2
(z a)
2
, which is
the negative of the previous answer. Summarizing
_
|z|=
[z a[
4
[dz[ =
_

_
2(
2
+[a[
2
)
(
2
[a[
2
)
3
if [a[ <

2(
2
+[a[
2
)
(
2
[a[
2
)
3
if [a[ > .
4.2.3.2. Prove that a function which is analytic in the whole plane and
satises an inequality [f(z)[ < [z[
n
for some n and all suciently large [z[
reduces to a polynomial.
Solution:
We assume f is globally analytic and there is C so [f(z)[ C[z[
n
for all
suciently large [z[. Let P(z) be the (n 1)
th
order Taylor polynomial of f(z)
at 0, so f(z)P(z) has a zero of order n at 0 and therefore f(z)P(z) = z
n
g(z)
for some globally analytic g(z). Since P is an order n 1 polynomial we have
Selected Homework Solutions 165
[P(z)[/[z[
n
1 for all suciently large [z[; using the hypothesis we get
[g(z)[
[f(z)[
[z[
n
+
[P(z)[
[z[
n
C + 1
for all suciently large [z[, whence continuity of g implies it is bounded and
Liouvilles theorem implies it is a constant a. Thus f(z) = az
n
+ P(z) is a
polynomial of degree at most n.
4.2.3.5. Show that the successive derivatives of an analytic function at a
point can never satisfy [f
(n)
(z)[ > n!n
n
. Formulate a sharper theorem of the
same kind.
Solution:
Suppose f is analytic at z
0
. Let g(z) = f(z) f(z
0
). If n 1 is an integer
and r > 0 is suciently small (so f is analytic on [z[ r), then by the Cauchy
formula

f
(n)
(z
0
)

g
(n)
(z
0
)

n!
2i
_
|zz0|=r
g(z)
(z z
0
)
n+1
dz

n!
r
n
max
|zz0|=r
[g(z)[
=
n!
r
n
max
|zz0|=r
[f(z) f(z
0
)[.
Putting r = 1/n we see that
n!n
n

f
(n)
(z
0
)

max
|zz0|=1/n
[f(z) f(z
0
)[ 0
as n . This is strictly stronger than the statement that

f
(n)
(z
0
)

> n!n
n
cannot occur for an unbounded sequence of n N.
4.3.2.1. If f(z) and g(z) have the algebraic orders h and k at z = a, show
that fg has the order h+k, f/g the order hk, and f +g an order which does
not exceed max(h, k).
Solution:
166 Homework Set 6
The algebraic order of f at a is dened to be that h such that
lim
za
[z a[

[f(z)[ =
_
_
_
< h
0 > h.
If f has order h and g has order k at z = a, then it is immediate that
[z a[

[f(z)g(z)[ =
for < h +k and = 0 for > h +k, so the order of f(z)g(z) is h +k. It is also
easy to see that the order of 1/g is k; combining these results we see that the
order of f(z)/g(z) is h k.
To see that the order of f + g cannot exceed maxh, k it suces to note
that [z a[

[f(z) +g(z)[ [z a[

[f(z)[ +[z a[

[g(z)[ 0 as z a, by the
triangle inequality.
4.3.2.3. Show that the functions e
z
, sin z, and cos z have essential singular-
ities at .
Solution:
The map e
z
takes any innite horizontal strip of width 2 to C 0, so it
takes any punctured neighborhood [z[ > R of to C 0. It follows that e
z
does not have a limit in

C as z , and therefore that the isolated singularity
of e
z
at is essential. Similar arguments apply to cos z and sin z. All we need
verify is that the image of [z[ > R cannot converge in

C as R , which is
true for cos because it maps all lines z = 2k + iy onto [0, ) R and for sin
because it maps the same lines onto the imaginary axis.
4.3.2.5. Prove that an isolated singularity of f(z) is removable as soon as
either Re(f(z)) or Im(f(z)) is bounded above or below. Hint: Apply a fractional
linear transformation.
Solution:
Suppose f has an isolated singularity at a, so f is analytic on 0 < [z
a[ < . If either Re f or Imf is bounded on this punctured disc, then the
image of the punctured disc cannot be dense in C, so the Casorati-Weierstrass
theorem implies the singularity cannot be essential. The singularity is therefore
removable or a pole. However, if it is a pole then f(z) = (z a)
k
g(z) for some
Selected Homework Solutions 167
k N and g analytic on [z a[ < with g(a) ,= 0. Then we can easily verify
that both Re f and Imf are unbounded on 0 < [z a[ < as follows. Take
z = a + re
i
, so f(z) = r
k
e
ik
g(z). For r > 0 so small that [g(z) g(a)[ <
[g(a)[/2 and chosen so e
ik
g(a) is real we have Re f(z) r
k
[g(a)[/2. A
similar argument works for the imaginary part. It follows that the singularity
cannot be a pole, so it must be removable.
4.3.2.6. Show that an isolated singularity of f(z) cannot be a pole of
exp(f(z)). Hint: f and e
f
cannot have a common pole (why?). Now apply
Theorem 9.
Solution:
Let f have an isolated singularity at a, so f is analytic on 0 < [z a[ < .
Consider g(z) = e
f(z)
, which also has an isolated singularity at a. Suppose the
singularity of g(z) is a pole. Then lim
za
[g(z)[ = , and [g(z)[ = e
Re f(z)
, so
lim
za
Re f(z) = . In particular, f(z) is unbounded on 0 < [z a[ < , so
the singularity of f is not removable. Also, f(z) omits a left half-plane, so the
Casorati-Weierstrass theorem implies the singularity of f cannot be essential.
Thus f has a pole at a. Reasoning as in the previous question we see that for
any suciently small r > 0, the image of any 0 < [z a[ < r under f covers
a set of the form [z[ > R. However the exponential map takes any horizontal
innite strip of height 2 to C0. Since a region of the form [z[ > R contains
such a horizontal strip we conclude that g maps any 0 < [z a[ < r to all of
C0, contradicting the assumption that g has a pole at a. We conclude that
if f(z) has an isolated singularity at a then e
f(z)
cannot have a pole at a.
10.7 Homework Set 7
4.3.3.1. Determine explicitly the largest disk about the origin whose image
under the mapping w = z
2
+z is one-to-one.
Solution:
Observe that f(0) = 0 and f

(0) ,= 0, so that f is injective on some neigh-


borhood of 0. Also f(z) is not injective on [z[ < 2, because f(0) = f(1) = 0.
It follows that R = supr : f is injective on [z[ < r is positive and nite.
168 Homework Set 7
One easy way to nd R is by direct computation. The roots of f(z) are
z =
1
2

_
1 + 4
2
2
,
so they are at opposite points of a diameter of a circle radius
s =

_
1 + 4
2
2

around
1
2
. One of these points has real part less than
1
2
, so is distance at
least
1
2
from 0. We conclude that no open disc [z[ < r with r
1
2
contains two
roots of f(z) . Moreover if r >
1
2
we may choose to make s so small that
[z
1
2
[ < s is inside [z[ < r, at which point there will be two roots of f(z)
in [z[ < r. This proves that R =
1
2
.
A more general approach is to rst recognize that an analytic f cannot
be injective on an open set containing a critical point. The reason is that if
f

(z
0
) = 0, then f(z) f(z
0
) has a zero of some order n 2 at z
0
. The
argument principle then says that if w is suciently close to f(z
0
), the function
f(z) w must have n simple roots in a neighborhood of z
0
, so f cannot be
injective. (See book, top of page 132.) For our problem, the critical point is at
2z + 1 = f

(z) = 0, so z =
1
2
and we discover R
1
2
.
Now if f is a polynomial, then we know that all critical points lie in the
convex hull of the roots (see Theorem 1 on page 29 of the book). So if we have a
polynomial f of degree n and a w so f(z)w has n roots (counting multiplicity)
in [z[ < r, then f also has a critical point in [z[ < r. In the situation we face,
with n = 2, we discover that if there are 2 roots of f(z) in [z[ < r then
r >
1
2
. Thus R
1
2
, and we conclude R =
1
2
.
It is interesting to think how, if at all, this idea could be generalized.
4.3.3.4. If f(z) is analytic at the origin and f

(0) ,= 0, prove the existence


of an analytic g(z) such that f(z
n
) = f(0) +g(z)
n
in a neighborhood of 0.
Solution:
We are given f analytic at 0 and f

(0) ,= 0. Then
f(z) f(0) = zh(z)
with h analytic at 0 and h(0) = f

(0) ,= 0. We may then take an open disc U


Selected Homework Solutions 169
around 0 which is so small that h is analytic on this disc and
[h(z) h(0)[ < [h(0)[
for z U. This condition on h implies log h(z) and hence k(z) = h(z)
1/n
are
well-dened (single-valued) analytic functions on U, from which
f(z
n
) = f(0) +z
n
h(z
n
) = f(0) +
_
zk(z
n
)
_
1/n
on U, so taking g(z) = zk(z
n
) completes the proof.
10.8 Homework Set 8
4.3.4.1. Show by use of

M(f(z) w
0
)
M
2
w
0
f(z)

R(z z
0
)
R
2
z
0
z

,
or directly, that [f(z)[ 1 for [z[ 1 implies
[f

(z)[
(1 [f(z)[
2
)

1
1 [z[
2
.
Solution:
Suppose [f(z)[ 1 on [z[ 1 and f analytic. Fix z
0
in the open unit disc
and let w
0
= f(z
0
). Using a composition of fractional linear transformations
and the Schwarz lemma, it is proved in the book that

f(z) w
0
1 w
0
f(z)

z z
0
1 z
0
z

. (10.1)
With a little rewriting we can extract f

(z
0
):
[f

(z
0
)[
1 [f(z)[
2
= lim
zz0

f(z) w
0
(z z
0
)(1 w
0
f(z))

lim
zz0

1
1 z
0
z

=
1
1 [z
0
[
2
(10.2)
and z
0
was an arbitrary point in the disc, so the estimate is valid for all [z
0
[ < 1.
170 Homework Set 8
4.3.4.2. If f(z) is analytic and Im(f(z)) 0 for Im(z) > 0, show that
[f(z) f(z
0
)[
[f(z) f(z
0
)[

[z z
0
[
[z z
0
[
and
[f

(z)[
Im(f(z))

1
Im(z)
.
Solution:
Suppose f is non-constant analytic and Imf(z) 0 if Imz 0. Composition
with a fractional linear map gives a map from the unit disc to itself to which
we can apply the Schwarz lemma. Specically, for any z
0
with Imz
0
> 0 we
set w
0
= f(z
0
) and have Imw
0
> 0 by the maximum principle. For in the
upper half-plane let g

(z) =
z
z
so g
1
z0
maps the unit disc to the upper half-
plane with 0 z
0
and g
w0
maps the upper half-plane to the unit disc with
w
0
0. Thus g
w0
f g
1
z0
takes the unit disc to itself and xes 0, whence
[g
w0
f g
1
z0
(z)[ [z[ by the Schwarz lemma, and so [g
w0
f(z)[ [g
z0
(z)[.
Substituting gives

f(z) f(z
0
)
f(z) +f(z
0
)

z z
0
z +z
0

. (10.3)
As in the previous question, we can obtain a bound for [f

(z
0
)[ by dividing
both sides of the previous equation by [z z
0
[ and sending z z
0
. Since
z +z = 2 Imz, we obtain
[f

(z
0
)[
Imf(z
0
)

1
Imz
0
(10.4)
valid at all points in the upper half-plane.
4.3.4.3. In Exercises 4.3.4.1 and 4.3.4.2, prove that equality implies that
f(z) is a linear transformation.
Solution:
Let
F(z) =
(f(z) w
0
)(1 z
0
z)
(z z
0
)(1 w
0
f(z))
.
Then F is analytic on the open unit disc except for a removable singularity at
z
0
, and (10.1) says [F(z)[ 1. If equality holds in (10.2) then [F(z
0
)[ = 1, so
the maximum principle implies F is constant of modulus 1, which we may write
as e
i
, [0, 2). Multiplying out gives that f(z) = h
1
w0
(e
i
h
z0
(w)) where
Selected Homework Solutions 171
h

=
z
1z
.
Similarly, if we let
G(z) =
(f(z) w
0
)(z +z
0
)
(z z
0
)(f(z) +w
0
)
then we obtain a function analytic on Imz > 0 except for the removable sin-
gularity at z
0
, and [G(z)[ 1 by (10.3). Equality in (10.4) implies G e
i
for some [0, 2) by the same maximum principle argument as before, and
inverting the maps we have f(w) = g
1
w0
(e
i
g
z0
(w)) with g as in the previous
exercise.
In either of these two cases we see that f is a composition of fractional linear
maps, so is fractional linear.
4.4.7.3. Show that the bounded regions determined by a closed curve are
simply connected, while the unbounded region is doubly connected.
Solution:
Let be a closed curve. Its complement is open, so the connected compo-
nents of the complement are open, and there are thus at most countably many
of them. Label them U
j

j=0
, and let U
0
be the unique unbounded one; it is
unique because compactness of implies the complement of some large disc is
in C , and this is a connected set so is contained in a single component.
For each j, let V
j
=

C U
j
and observe
V
j
=
_
k=j
U
j
.
As these sets are connected, each must lie in exactly one component of V
j
.
However connectivity of U
k
implies connectivity of its closure, which intersects
, so all U
k
and must lie in a single component of V
j
. Moreover if j ,= 0 then
U
0
V
j
and is its closure, so is in this same component of V
j
. We conclude
that if j ,= 0 then V
j
has only one component, so U
j
is simply connected by
denition. Also V
0
has at most two components, one being and the other
_
j1
U
j
.
Taking r so [z[ = r does not intersect , we see that the disjoint open sets [z[ > r
and [z[ < r separate V
0
into these two components, so V
0
is doubly connected.
172 Homework Set 8
4.4.7.5. Show that a single-valued analytic branch of

1 z
2
can be de-
ned in any region such that the points 1 are in the same component of the
complement. What are the possible values of
_
dz

1 z
2
over a closed curve in the region?
Solution:
Let be a region not intersecting a connected set E with 1 E. If (z)
is a globally analytic function such that
f(z) =
1 z
2
((z))
2
maps E to a connected region containing 0, then f has a well-dened loga-
rithm and therefore well-dened roots on on Cf(E). Provided f() does not
intersect f(E) we may dene a single-valued analytic function

1 z
2
on by
_
1 z
2
= (z)

1 z
2
((z))
2
and it is a legitimate branch of the square root because squaring both sides
leads to an equality.
It remains to nd such a (z), but writing 1 z
2
= (1 z)(1 +z) immedi-
ately suggests (z) = (1 + z), as then f(z) =
1z
1+z
, which is a fractional linear
transformation with 1 and 1 0, from which f(E) is a connected set
containing 0 and . Our denition is then
_
1 z
2
= (1 +z)
_
1 z
1 +z
.
Now consider the integral
_

dz

1 z
2
=
_

1
(1 +z)
_
1z
1+z
dz
where is a closed curve in a component of C E. The integrand has no
singularities in this component, so if does not wind around any point of E
then the integral is zero. In particular if E then cannot wind around E,
Selected Homework Solutions 173
so the integral is zero in this case.
We therefore suppose that E is bounded and take r > 0 so large that E
z : [z[ < r. In this case the winding number n(, z) is a constant 2iN on E
and is homologous in the unbounded component of C E to N
r
, where
r
is the circle of radius r around 1. The Cauchy theorem then implies
_

dz

1 z
2
= N
_
r
1
(1 +z)
_
1z
1+z
dz
= N
_
2
0
ire
i
d
re
i
_
2
re
i
1
= iN
_
2
0
d
_
2
re
i
1
It is tempting at this point to attempt to use the residue theorem, but that
theorem is valid only for isolated singularities, and our square root has singu-
larities along a connected set joining 1. Instead we observe that we may take
r without changing the value of the integral. We nd that the integrand
converges to 1/

1, which is one of i (we cannot determine which without


knowing more about the set E; draw some pictures to see why). Therefore the
possible values of the integral are 2N, or any element of 2Z.
4.5.2.1. How many roots does the equation z
7
2z
6
+6z
3
z +1 = 0 have
in the disk [z[ < 1? Hint: Look for the biggest term when [z[ = 1 and apply
Rouches theorem.
Solution:
Let f(z) = z
7
2z
6
z +1. Then [f(z)[ 5 on [z[ = 1, so Rouches theorem
implies 6z
3
+f(z) has the same number of roots as 6z
3
in the unit disc, namely
three.
4.5.2.2. How many roots of the equation z
4
6z +3 = 0 have their modulus
between 1 and 2?
Solution:
We use Rouches theorem twice. For [z[ = 2, [z
4
[ = 16 > [ 6z + 3[, so
z
4
6z + 3 has 4 roots in [z[ 2. For [z[ = 1, [6z[ = 6 > [z
4
+ 3[, so there is
one root in [z[ 1. We conclude that there are 3 roots in 1 < [z[ < 2.
174 Homework Set 9
10.9 Homework Set 9
4.5.3.1. Find the poles and residues of the following functions:
(a)
1
z
2
+ 5z + 6
, (b)
1
(z
2
1)
2
, (c)
1
sin z
, (d) cot z,
(e)
1
sin
2
z
, (f)
1
z
m
(1 z)
n
, m, n N.
Solution:
(a)
f(z) =
1
z
2
+ 5z + 6
=
1
(z + 2)(z + 3)
which has
a pole of order 1 at z = 2 with residue lim
z2
(z + 2)f(z) = 1
a pole of order 1 at z = 3 with residue lim
z3
(z + 3)f(z) = 1
(b)
f(z) =
1
(z
2
1)
2
=
1
(z 1)
2
(z + 1)
2
which has
a pole of order 2 at z = 1 with residue lim
z1
d
dz
(z 1)
2
f(z) =
1
4
a pole of order 2 at z = 1 with residue lim
z1
d
dz
(z + 1)
2
f(z) =
1
4
(c) f(z) =
1
sin z
has poles at the zeros of sin z, so at the points k, k
Z. These zeros are simple, because f

(z) = cos z = 1 at these points, and


consequently the poles are simple. The residue at k may be computed by
LHopitals rule
lim
zk
z k
sin z
= lim
zk
1
cos z
= (1)
k
so that f(z) has simple poles with residue (1)
k
at each k, k Z.
(d) As cos z is entire, f(z) = cot z =
cos z
sin z
can only have poles at the zeros of
sin z, meaning the points z = k, k Z. Since cos z ,= 0 at these points, there
is a pole at each such point, and since the zeros of sin z are simple the poles are
Selected Homework Solutions 175
also simple. The residue at k is
lim
zk
(z k) cos z
sin z
= cos(k) lim
zk
1
cos z
= 1
so cot z has simple poles with residue 1 at each k, k Z.
(e) f(z) =
1
sin
2
z
has poles at each of the zeros z = k, k Z of sin z.
These zeros are order 1, so the zeros of sin
2
z are order 2. The residues may be
computed using LHopital
lim
zk
d
dz
_
(z k)
2
sin
2
z
_
= lim
zk
2(z k) sin z 2(z k)
2
cos z
sin
3
z
= 2 lim
zk
_
z k
sin z
_
lim
zk
_
sin z (z k) cos z
sin
2
z
_
= 2 lim
zk
_
cos z cos z + (z k) sin z
2 sin z
_
= lim
zk
(z k) = 0.
We determine that f(z) has poles of order 2 at each of the points k, k Z and
has residue zero at each pole.
(f) f(z) = z
m
(1 z)
n
, m, n N has a pole of order m at 0 and a pole of
order n at 1. We may compute the residue at 0 using
lim
z0
1
(m1)!
d
m1
dz
m1
(1 z)
n
=
1
(m1)!
(n +m2)!
(n 1)!
(1 0)
n(m1)
=
_
(n 1) + (m1)
m1
_
=
_
(n 1) + (m1)
n 1
_
The residue at 1 may be computed the same way
lim
z1
1
(n 1)!
d
n1
dz
n1
(z 1)
n
f(z) =
1
(n 1)!
lim
z1
d
n1
dz
n1
(1)
n
z
m
= (1)
n
(1)
n1
(m+n 2)!
(n 1)!(m1)!
(1)
m(n1)
=
_
(n 1) + (m1)
m1
_
=
_
(n 1) + (m1)
n 1
_
so we conclude that f has a pole of order m at zero with residue
_
(n1)+(m1)
n1
_
and a pole of order n at 1 with residue
_
(n1)+(m1)
n1
_
. It is worth noting that
this is consistent with exercise 4.2.3.1(b).
176 Homework Set 9
4.5.3.3. Evaluate the following integrals by the method of residues:
(a)
_
/2
0
dx
a + sin
2
x
, [a[ > 1, (b)
_

0
x
2
dx
x
4
+ 5x
2
+ 6
,
(c)
_

x
2
x + 2
x
4
+ 10x
2
+ 9
dx, (d)
_

0
x
2
dx
(x
2
+a
2
)
2
, a R,
(e)
_

0
cos x
x
2
+a
2
dx, a R, (f)
_

0
xsin x
x
2
+a
2
dx, a R,
(g)
_

0
x
1/3
1 +x
2
dx, (h)
_

0
log x
1 +x
2
dx,
(i)
_

0
log(1 +x
2
)
dx
x
1+
, 0 < < 2.
Try integration by parts in part (i).
Solution:
There are a few things we will use repeatedly in computing these integrals.
The curve
R
will be the semicircle [z[ = R in the upper half-plane, with the
usual (increasing angle) orientation. For R, S, T (0, ) we also let

1
= S +iy : 0 y T,

2
= x +iT : R x S,

3
= R +iy : 0 y T,
oriented such that
1
+
2
+
3
and the interval [R, S] R form a positively
oriented closed curve. We will frequently use that (A) if the integrand f(x) is
bounded by [x[
2
as [x[ then
_

= lim
R
_
R
R
f(x) dx,
and (B) if the integrand f(z) is bounded by [z[
2
as [z[ then
lim
R
_

R
f(z) dz = 0.
Selected Homework Solutions 177
4.5.3.3(a). Note that sin
2
x = sin
2
(x) = sin
2
( x) = sin
2
( +x) implies
_
/2
0
dx
a + sin
2
x
=
1
4
_
2
0
dx
a + sin
2
x
and this may be seen as an integral on the unit circle with respect to the angle
dx = dz/iz where sin z = (z z
1
)/2i. Thus
_
/2
0
dx
a + sin
2
x
=
1
4
_
|z|=1
1
a +
_
(z z
1
)/2i
_
2
dz
iz
.
We simplify
_
(z z
1
)/2i
_
2
= (2z)
2
(z
4
2z
2
+ 1) and nd the integrand
becomes
4iz
z
4
(2 + 4a)z
2
+ 1
.
At this point we can make our lives a little easier by making the substitution
w = z
2
. Notice that when z winds once around the unit circle, w winds around
twice. We therefore nd
_
/2
0
dx
a + sin
2
x
=
1
2
_
|w|=1
2i
w
2
(2 + 4a)w + 1
dw = i
_
|w|=1
dw
w
2
(2 + 4a)w + 1
.
Now we would like to say that the poles of w
2
(2 + 4a)w + 1 are at
w

= 1 + 2a 2
_
a
2
+a
by the quadratic formula, but this requires that we make sense of the square
root. Fortunately, [a[ > 1 by hypothesis, so [1/a[ < 1 and
_
1 + (1/a) is well-
dened. We may therefore dene an analytic branch of

a
2
+a by

a
2
+a =
a
_
1 + (1/a), obtaining w

= 1 +2a 2a
_
1 + (1/a). By construction, each of
w

is a branch of the inverse of the map w (w + w


1
)
2
/4, evaluated at a.
Since w (w+w
1
)
2
/4 takes the unit circle to the interval [1, 1], and [a[ > 1,
we see that [w

[ ,= 1 (this is important for applying the residue theorem). It


also follows that only one of w

can lie inside the unit disc, and that which one
does so is independent of a. Taking a (1, ) we readily see [w

[ < 1 and
[w
+
[ > 1, so this must be true for all a. Thus the result of the integration can
178 Homework Set 9
be computed from the residue at the simple pole w

, which has value


1
w

w
+
=
1
4a
_
1 + (1/a)
.
Finally
_
/2
0
dx
a + sin
2
x
=
2i
2
(4a
_
1 + (1/a))
=

2a
_
1 + (1/a)
.
4.5.3.3(b). Using that the integrand is even and (A), then the residue
theorem, and then (B)
_

0
x
2
dx
x
4
+ 5x
2
+ 6
=
1
2
_

z
2
dz
z
4
+ 5z
2
+ 6
=
1
2
lim
R
_
R
R
z
2
dz
z
4
+ 5z
2
+ 6
= i

j
Res
zj

1
2
lim
R
_

R
z
2
dz
z
4
+ 5z
2
+ 6
= i

j
Res
zj
z
2
z
4
+ 5z
2
+ 6
where the sum is over residues in the upper half-plane. Now
z
4
+ 5z
2
+ 6 = (z
2
+ 2)(z
2
+ 3),
so the integrand has simple poles at z = i

2 and z = i

3. We have
Res(f, i

2) =
2
(2

2i)(1)
,
and
Res(f, i

3) =
3
(1)(2

3i)
,
so the result is
_

0
x
2
dx
x
4
+ 5x
2
+ 6
= i
_

2
2i
+

3
2i
_
= (

2)

2
.
Selected Homework Solutions 179
4.5.3.3(c). By (A), the residue theorem, and (B)
_

x
2
x + 2
x
4
+ 10x
2
+ 9
dx = lim
R
_
R
R
z
2
z + 2
z
4
+ 10z
2
+ 9
dz
= 2i

j
Res
zj
lim
R
_

R
z
2
z + 2
z
4
+ 10z
2
+ 9
dz
= 2i

j
Res
zj
z
2
z + 2
z
4
+ 10z
2
+ 9
where the sum is over residues in the upper half-plane. The zeros of
z
4
+ 10z
2
+ 9 = (z
2
+ 1)(z
2
+ 9)
are at i and 3i; each produces a simple pole in the integrand, and there are
no others. The residue at i is (1 i +2)/(2i)(1 +9) and at 3i is (9 3i +
2)/(9 + 1)(6i), so the result is
_

x
2
x + 2
x
4
+ 10x
2
+ 9
dx = 2i
_
1 i
16i
+
7 + 3i
48i
_
= (3 3i + 7 + 3i)

24
=
5
12
.
4.5.3.3(d). Using that the integrand is even and (A), then the residue
theorem, and then (B)
_

0
x
2
dx
(x
2
+a
2
)
3
=
1
2
_

z
2
dz
(z
2
+a
2
)
3
=
1
2
lim
R
_
R
R
z
2
dz
(z
2
+a
2
)
3
= i

j
Res
zj

1
2
lim
R
_

R
z
2
dz
(z
2
+a
2
)
3
= i

j
Res
zj
z
2
(z
2
+a
2
)
3
where the sum is over residues in the upper half-plane. Factoring
(z
2
+a
2
)
3
= (z +ai)
3
(z ai)
3
, a R,
we see that there is a single pole of order 3 in the upper half-plane, at i[a[. The
180 Homework Set 9
residue there is
lim
z|a|i
d
2
dz
2
z
2
(z +[a[i)
3
=
1
8[a[
3
i
so that the result is
_

0
x
2
dx
(x
2
+a
2
)
3
=

8[a[
3
4.5.3.3(e). Using that the integrand is even and (A), that cos z = Re e
iz
,
then the residue theorem for R = S and T suciently large,
_

0
cos x dx
x
2
+a
2
=
1
2
_

cos z dz
z
2
+a
2
=
1
2
lim
R
Re
_
R
R
e
iz
dz
z
2
+a
2
= i

j
Res
zj
e
iz
z
2
+a
2

1
2
lim
R,T
Re
__
1
e
iz
dz
z
2
+a
2
_
2
e
iz
dz
z
2
+a
2
+
_
3
e
iz
dz
z
2
+a
2
_
where the sum is over residues in the upper half-plane. However the integrand
is bounded by a constant multiple of e
y
/[z[
2
for z = x +iy and [z[ suciently
large. Writing f(z) for the integrand, and taking R = S and T large enough we
nd that

_
1
f(z) dz

1
R
2
_
T
0
e
y
dy
1
R
2
,
and similarly for
3
. Now on
2
we have that (z
2
+ a
2
)
1
is integrable (with
integral bounded by constant C) if T is large enough, and therefore

_
2
f(z) dz

Ce
T
.
Sending R and T to we nd
_

0
cos x dx
x
2
+a
2
= Re i Res
|a|i
f(z) =

2[a[
where at the last step we computed that z
2
+ a
2
= (z + ai)(z ai), has one
simple pole in the upper half-plane, at i[a[, with residue
lim
z|a|i
cos z
z +[a[i
=
cos a
2[a[i
.
Selected Homework Solutions 181
4.5.3.3(f ). We use that the integrand is even and
xsin x
x
2
+a
2
= Im
_
ze
iz
z
2
+a
2
_
.
Taking R, S, T large enough that the curve (R, S)
1

2

3
encloses the
simple pole at [a[i, where the residue is
lim
z|a|i
ze
iz
z +[a[i
=
[a[ie
|a|
2[a[i
=
e
|a|
2
we obtain
_

0
xsin x
x
2
+a
2
dx =
1
2
lim
R,S
Im
_
S
R
ze
iz
z
2
+a
2
dz
= Imi
e
|a|
2
lim
R,S
Im
__
1
ze
iz
z
2
+a
2
dz +
_
2
ze
iz
z
2
+a
2
dz +
_
3
ze
iz
z
2
+a
2
dz
_
valid for all suciently large T. However, the integrand f(z) satises
[f(z)[
[z[e
y
[z[
2
[a[
2
for z = x +iy. The integral for
1
can be bounded by
R
R
2
[a[
2
_
T
0
e
y
dy =
R
R
2
[a[
2
and similarly that for
3
can be bounded by
S
S
2
[a[
2
.
The integral for
2
can be bounded by
Se
T
S
2
[a[
2
(R +S).
If we rst send T so the
2
integral goes to 0, and then send R, S
we nd that they make no contribution to the result, and therefore
_

0
xsin x
x
2
+a
2
dx = Imi
e
|a|
2
=

2e
|a|
.
182 Homework Set 9
4.5.3.3(g). We will do this for general (1, 1), as it will be useful later.
Take > 0, > 0, and R > 2. Let

= re
i
, r (, R)
be rays at angle , and also take arcs

R
= Re
i
: (, 2 )
and

= e
i
: (, 2 ).
Let z

be a branch on C[0, ), so it is well-dened and analytic in a simply


connected neighborhood of the closed curve
+
+
R

. Provided
and are suciently small, this curve winds once around the simple poles of
f(z) = z

(1 + z
2
)
1
, which are at i, and where there are residues i

/2i and
(i)

/ 2i respectively.
_
++
R

z
2
+ 1
dz = 2i(i

(i)

)/2i = (e
i/2
e
i3/2
).
Now on
+
we have z

= r

e
i
, while on

, z

= r

e
i(2)
. It follows that
lim 0
_
+
z

z
2
+ 1
dz = (1 e
i2
)
_
R

x
2
+ 1
dx.
At the same time, we see that on

the integrand has the bound [f(z)[ 2

,
and the length of the curve is less than 2, so the integral is bounded by
4
1+
0 as 0, provided > 1. On
R
we have
[f(z)[
R

R
2
1
,
and the curve has length less than 2R, so the integral is bounded by
2R
1+
R
2
1
0
Selected Homework Solutions 183
as R provided < 1. We conclude that if (1, 1) then
_

0
x

x
2
+ 1
dx = lim
0,R,0
1
1 e
i2
_
++
R

z
2
+ 1
dz
=
e
i/2
e
i3/2
1 e
i2
=
e
i
(e
i/2
e
i/2
)
e
i
(e
i
e
i
)
=
sin(/2)
sin
=
sin(/2)
2 sin(/2) cos(/2)
=

2
sec
_

2
_
.
In the special case =
1
3
, we have sin(/6)/ sin(/3) = 1/

3, so that
_

0
x
1/3
x
2
+ 1
dx =

3
.
4.5.3.3(h). For this problem, let
+
= (, R) R and

= (R, ) R,

and
R
be the semicircles of radius and R (respectively) in the upper half-
plane. Dene log z to be the branch of the logarithm on the complement of the
negative imaginary axis. Taking
+
+
R

arranged to form a curve


winding once around the simple pole of
log z
(z
2
+1)
at z = i, we nd from the residue
theorem that
_
++
R
+
log z
z
2
+ 1
dz = 2i(log i)/2i =
i
2
2
.
The computations showing that the contributions from

and
R
vanish in the
limit are essentially the same as in exercise 4.5.3.3.g. All that is dierent is we
use the bound [ log z[ (log [z[ +2). Since log z is log [z[ on
+
and log [z[ +i
on

we nd
_
++
log z
z
2
+ 1
dz =
_
R

log x
x
2
+ 1
dx +
_

R
log x +i
x
2
+ 1
dx
= 2
_
R

log x
x
2
+ 1
dx +
_
R

1
x
2
+ 1
dx.
184 Homework Set 9
Combining these facts we see
i
2
2
= lim
0,R
_
++
R
+
log z
z
2
+ 1
dz
= lim
0,R
_
++
log z
z
2
+ 1
dz
= 2
_

0
log x
x
2
+ 1
dx +
_

0
i
x
2
+ 1
dx
= 2
_

0
log x
x
2
+ 1
dx +
i
2
2
so that
_

0
log x
x
2
+ 1
dx = 0.
4.5.3.3(i). Let us rst observe that f(x) = x
(1)
log(1+x
2
) is integrable
on [0, ) because it is bounded by C

x
1(/2)
as x and > 0, while as
x 0 one has [ log(1 + x
2
)[ 2x
2
so [f(x)[ x
1
and < 2. It follows that
we can write the integral as a limit and can integrate by parts
_

0
log(1 +x
2
)
x
1+
dx = lim
R
_
R
1
R
log(1 +x
2
)
x
1+
dx
= lim
R
_
x

log(1 +x
2
)
_
R
1
R
+
1

lim
R
_
R
1
R
2x
1
1 +x
2
dx.
We observe that as R , R

log(1 +R
2
) 0, and also
[R

log(1 +R
2
)[ 2R
2
0,
so the boundary term from the integration makes no contribution in the limit.
The remaining term may be dealt with by the computation in 4.5.3.3.g. Indeed,
from that problem with = (1 ) (1, 1), we have
_

0
log(1 +x
2
)
x
1+
dx =
1

_

0
2x
1
1 +x
2
dx =

sec
_
(1 )
2
_
=

csc
_

2
_
4.5.3.4. Compute
_
|z|=
[dz[
[z a[
2
, [a[ , = .
Selected Homework Solutions 185
Hint: Use zz =
2
to convert the integral to a line integral of a rational function.
Solution:
Parameterizing [z[ = by z = e
i
we have dz = iz d and [dz[ = d, so
[dz[ = dz/iz. Also [z a[
2
= (z a)(z a) = (z a)(

2
z
a). Hence we nd
_
|z|=
[dz[
[z a[
2
=
_
|z|=

iz(z a)(

2
z
a)
dz =
_
|z|=

i(z a)(
2
az)
dz
which can be computed by the residue theorem. There are simple poles at a
and
2
/a. By hypothesis, [a[ , = ; if [a[ < then a is inside [z[ = and
2
/a is
not, and the reverse is true if [a[ > .
The residue at z = a is

i(
2
|a|
2
)
and that at
2
/a is

i(
2
|a|
2
)
. We conclude
from the residue theorem that
_
|z|=
[dz[
[z a[
2
=
_

_
2

2
[a[
2
if [a[ <
2

2
[a[
2
if [a[ >
=
2

2
[a[
2

.
10.10 Homework Set 10
4.6.2.2. Suppose that f(z) is analytic in the annulus A(0, r
1
, r
2
) and contin-
uous on the closed annulus. If M(r) denotes the maximum of [f(z)[ for [z[ = r,
show that
M(r) M(r
1
)

M(r
2
)
1
(10.5)
where
=
log(r
2
/r)
log(r
2
/r
1
)
(Hadamards three-circle theorem). Discuss cases of equality. Hint: Apply the
maximum principle to a linear combination of log [f(z)[ and log [z[.
Solution:
If M(r) = 0 for some r > 0 then f vanishes on a set containing a limit
point, so f 0 and the result is trivial. Hence there is no loss of generality in
assuming M(r) > 0 for r > 0, in which case the statement
M(r) M(r
1
)

M(r
2
)
(1)
186 Homework Set 10
for
=
log(r
2
/r)
log(r
2
/r
1
)
is equivalent to
log M(r) log M(r
1
) + (1 ) log M(r
2
)
=
log r
2
log r
log r
2
log r
1
log M(r
1
) +
log r log r
1
log r
2
log r
1
log M(r
2
)
which is the same as
_
log r
2
log r
1
_
log M(r)
log r
2
log M(r
1
) log r
1
log M(r
2
) +
_
log M(r
2
) log M(r
1
)
_
log r
or
_
log M(r
1
) log M(r
2
)
_
log r +
_
log r
2
log r
1
_
log M(r)
log r
2
log M(r
1
) log r
1
log M(r
2
)
and it is this that we will prove.
It is suggested in the book that we apply the maximum principle (for har-
monic functions) to a linear combination of log [z[ + log [f(z)[. Of course we
cannot do this directly if f has zeros, because log [f(z)[ is not harmonic in any
neighborhood of a zero of f (in fact it is subharmonic, and there is still a max-
imum principle for subharmonic functions, but we have not proved that). We
will therefore need to do something about points where f is zero, but let us
begin by assuming that no such points exist.
If f is analytic on the annulus 0 < r
1
< [z[ < r
2
then Alog [z[ +Blog [f(z)[
is harmonic there, and the maximum principle for harmonic functions implies
that the maximum occurs on the boundary. We obtain
Alog r +Blog M(r) = max
|z|=r
_
Alog [z[ +Blog [f(z)[
_
(10.6)
max (Alog r
1
+Blog M(r
1
), Alog r
2
+Blog M(r
2
))
Taking A = log M(r
1
)log M(r
2
) and B = log r
2
log r
1
we nd that the terms
Selected Homework Solutions 187
on the right are both equal to log r
2
log M(r
1
) log r
1
log M(r
2
). Thus
log r
2
log M(r
1
) log r
1
log M(r
2
)

_
log M(r
1
) log M(r
2
)
_
log r +
_
log r
2
log r
1
_
log M(r)
which is what we needed to prove.
Now we deal with the points z
j
where f(z) = 0. Such points are can
accumulate only at the boundary. Suppose that around each we place a small
disc of radius
j
(small enough that it is inside the annulus), and delete these
discs from our domain. Then (10.6) must be modied so that for each j there is a
term on the right side corresponding to the maximum of Alog [z[ +Blog [f(z)[
on the new boundary circle [z z
j
[ =
j
. However Blog [f(z)[ as
z z
j
, so we may choose
j
so small that this new term is less than the right
side of (10.6), and therefore need not be included. It follows that (10.6) is still
valid when f has zeros, and therefore the result holds for general f.
Note that there is a degenerate case we did not consider, namely r
1
= 0. In
this situation one should interpret the formula for as corresponding to = 0,
whereupon the result follows directly from the usual maximum principle for the
harmonic function [f(z)[.
5.1.1.1. Use Taylors theorem applied to a branch of log(1 +
z
n
), prove that
lim
n
_
1 +
z
n
_
n
= e
z
(10.7)
uniformly on compact sets.
Solution:
Let K C be compact and M = max
zK
[z[. Observe that for n > M we
have

z
n

< 1 when z K. The principal branch of the logarithm is well-dened


on w : [1 + w[ < 1, so we conclude log
_
1 +
z
n
_
is well-dened on K for all
n > M and has Taylor expansion
log
_
1 +
z
n
_
=

j=1
(1)
j+1
j
_
z
n
_
j
.
The series is readily seen to be convergent on

1 +
z
n

< 1, thus uniformly


convergent on compact subsets of this region, and in particular on K for n > M.
188 Homework Set 10
Uniformity of the convergence implies we can exchange the limits in
lim
n
nlog
_
1 +
z
n
_
=

j=1
lim
n
(1)
j+1
j
z
j
n
j1
= z
Exponentiating both sides and using continuity of the exponential we get that
e
z
= lim
n
exp nlog
_
1 +
z
n
_
= lim
n
_
1 +
z
n
_
n
uniformly on K, and since K was arbitrary the convergence is uniform on all
compact sets in C.
5.1.2.3. Develop log(
sin z
z
) in powers of z up to the term z
6
.
Solution:
We wish to develop log
_
sin z
z
_
around 0 up to terms of order z
6
. Since sin z
has a simple zero at 0 the function
sin z
z
has a removable singularity at 0 and
its extension (which is equal to 1 at 0) is entire. It is helpful to recall the series
for sin z and divide by z to obtain a series convergent uniformly on all compact
sets to
sin z
z
sin z
z
=
1
z

j=0
(1)
j
z
2j+1
(2j + 1)!
=

j=0
(1)
j
z
2j
(2j + 1)!
= 1 +

j=1
(1)
j
z
2j
(2j + 1)!
.
Next we may compose with the series

k=1
(1)
k+1
k
w
k
for log(1 +w), which is convergent for [w[ < 1. This amounts to setting
w =

j=1
(1)
j
z
2j
(2j + 1)!
,
which we note satises [w[ < 1 on a neighborhood of 0. Observe that since the
lead z-term is z
2
it suces to consider the 3
rd
-order polynomial in w. We have
Selected Homework Solutions 189
log

sin z
z

sin z
z
1

1
2

sin z
z
1

2
+
1
3

sin z
z
1

3
+ [z
8
]
=
3
X
j=1
(1)
j
z
2j
(2j + 1)!

1
2

2
X
j=1
(1)
j
z
2j
(2j + 1)!
!
2
+
1
3

1
X
j=1
(1)
j
z
2j
(2j + 1)!
!
3
+ [z
8
]
=
z
2
3!
+
z
4
5!

z
6
7!

1
2

z
2
3!
+
z
4
5!

2
+
1
3

z
2
3!

3
+ [z
8
]
=

1
3!

z
2
+

1
5!

1
2(3!)
2

z
4
+

1
7!
+
1
(3!)(5!)

1
3(3!)
3

z
6
+ [z
8
]
=
z
3
3!
+
(3 5)z
4
2
3
3
2
5
+
(9 + 63 70)z
6
2
4
3
4
5 7
+ [z
8
]
=
z
3
2 3

z
4
2
2
3
2
5

z
6
3
4
5 7
+ [z
8
]
CHAPTER 11
Preliminary Exams
11.1 Syllabus
Holomorphic Functions
Statement of the Jordan curve theorem and the notion of simple rectiable
curves.
The Riemann sphere (1.3 on page 5).
The Cauchy-Riemann equations (page 11).
Power series and the circle of convergence (see Theorem 2.22 on page 20).
Linear fractional (Mobius) transformations (4.1 and conformal mapping.
Integration Theory
Integration along simple rectiable positively oriented curves (5.1 on page
57).
The Cauchy-Goursat theorem (page 64).
The Cauchy integral formula (page 71).
Moreras theorem (page 76) and the maximum principle (page 85).
191
192 Syllabus
Principle of the argument (page 93), winding numbers and Rouches the-
orem (page 93).
The residue theorem (page 93) and its use in evaluating real Riemann
integrals.
Representation Theorems
Taylor and Laurent series.
The maximum modulus theorem, Liouvilles theorem (page 76), and the
fundamental theorem of algebra (page 76).
Singularities.
Meromorphic functions.
Harmonic Functions
The mean value theorem (page 109, the maximum principle.
Their relationship to holomorphic (i.e., complex analytic) functions.
Harmonic conjugates (7.1 on page 103).
Miscellaneous
The inverse function theorem.
The Schwarz lemma (page 85).
The Schwarz reection principle.
Normal families (9.2 on page 123).
The Riemann mapping theorem (page 129).
References
[Ahl79], [Car95], [Con73], [Hil59], [Kno52], and [Rem91]
Preliminary Exams 193
11.2 January 2009
1. (a) Is there a holomorphic function f such that it maps D into itself,
f(0) = 0 and f(i/4) = i/3? Explain.
(b) Is there a holomorphic function f maps z : Re(z) > 0 into itself,
f(3) = 3 and f(9) = 6? Explain.
2. Show that there is no one-to-one holomorphic function D 0 onto the
annulus A = z : a < [z[ < b, where b > a > 0.
3. How many zeros counting multiplicities does e
z
2
4z
2
have in D?
4. Evaluate and justify your answer:
_

0
x
2m
1 +x
2n
dx, where m, n N 0, m < n.
5. Suppose f is entire, and [f(z)[ > 1 for [z[ > 1. Prove that f(z) is a
polynomial.
6. Consider a family T of function with domain D satisfying
f(z) =

n0
a
n
z
n
with

n0
[a
n
[ < 1.
Prove that any f T is holomorphic in D. Show that T is a normal
family.
194 August 2008
11.3 August 2008
1. Suppose f is entire, f(0) = 3 + 4i, and [f(z)[ 5 in D. Find f

(0).
2. Find a conformal map (an explicit formula) of the rst quadrant
z C : Im(z) > 0, Re(z) > 0
onto D. Is it unique? Can you give some conditions to ensure that this
conformal map is unique?
3. How many zeros counting multiplicities does z
5
+3z
3
+7 have in B(0, 2)?
4. Evaluate and justify your answer.
(a)
_

cos x
x
4
+ 1
dx. (b)
_
2
0
e
e
i
d.
5. Suppose that f is holomorphic in D such that
[f(z)[
1
1 [z[
, z D.
Prove that
[f

(z)[
4
(1 [z[)
2
, z D.
6. Consider a family of functions in Hol(D), the set of holomorphic func-
tions on the open unit disk, which are uniformly bounded at 0 and whose
derivatives are uniformly bounded on compact subsets of D. Show that
this family is normal. If the condition of uniform boundedness at 0 is
removed, what happens then? Explain.
Preliminary Exams 195
11.4 August 2007
1. (a) Find a conformal map from the set S = z : Im(z) > 0, Re(z) > 0
onto the open unit disk D such that 1 +i is mapped into 0.
(b) Find all the maps that satisfy (a), and prove that there are no others.
2. State and prove the Fundamental Theorem of Algebra.
3. How many zeros of the polynomial 2z
4
z
3
+ 5z
2
10z + 1 lie in the set
z : 1 < [z[ < 3? (Justify your answer.)
4. Evaluate the integral
_

sin x
x
3
+x
dx
using the Residue theorem (justify why it can be used). [Hint. The func-
tion
sin x
x
, for x R, is the imaginary part of an analytic function which
is bounded in the upper half-plane.]
5. Prove that if f is analytic in the open unit disk D, then
[f(w) f(0) wf

(0)[ [w[
2
sup
zD
[f(z) f(0) zf

(0)[,
for all w D.
6. Find all the entire functions f that satisfy
sup
zC
z=0,
f(z)=0

z log [z[
f(z)

< ,
and prove that there are no others.
7. Suppose that T is a family of analytic functions on the open unit disk
such that f(0) = 1 for all f T. Prove that if Im(f)
fF
is a normal
family, then T is also a normal family.
196 August 2006
11.5 August 2006
1. Suppose f is a nonconstant entire function such that (f f)(z) = f(z) for
all z. Prove that f must be the identity function.
2. Suppose f is entire, f(0) = 0 and [f(z)[ e
1/|z|
for all z ,= 0. Prove that
f is identically 0.
3. Suppose for each n that f
n
is a bounded continuous R-valued function on
the unit circle z : [z[ = 1. Suppose for each n that u
n
is a function that
is continuous on the closed unit disk z : [z[ 1, is harmonic in the open
unit disk z : [z[ < 1, and agrees with f
n
on the unit circle. Show that
f
n
is an equicontinuous family on the unit circle if and only if u
n
is
an equicontinuous family on the closed unit disk.
4. Use residues to evaluate the denite integral
_

x
2
(x
2
+ 1)
2
dx.
5. Let = z = x +iy : 0 < y < 1, x > 0. Find a conformal mapping of
onto the open unit disk.
6. Suppose that for each n the function f
n
is analytic in the open unit disk
D, [f
n
(0)[ 1, and for each r < 1 satises
_
|z|=r
[f
n
(z)[
2
[dz[ 1.
Show that every subsequence of f
n
has a further subsequence which
converges to a nite analytic function uniformly on each compact subset
of an open unit disk.
7. Suppose for each n the function f
n
is analytic on the open unit disk D
and has exactly one zero in D. Suppose the sequence f
n
converges to f
uniformly on each compact subset of the unit disk.
(a) Show that either f is identically zero on D or else has at most one
zero in D.
(b) Given an example of a sequence f
n
where the limit function has
no zeros in D.
Preliminary Exams 197
11.6 August 2005
1. (a) Prove the Minimum Principle for harmonic functions; i.e., show that
if u is harmonic in a region and u attains a minimum at a point
z
0
, then u is constant.
(b) Suppose f is analytic in the unit disk D and continuous in D, and
f(z) is real for [z[ = 1. Show that f is constant.
2. Let f and g be analytic and non-zero in a connected open set . Suppose
also that there exists a sequence of complex numbers z
n
so that
z
n
p and for all positive integers n,
f

(z
n
)
f(z
n
)
=
g

(z
n
)
g(z
n
)
.
Show that there is a constant c so that g = cf.
3. Show that if f is analytic in D and [f(z)[ 1 in D, then for all z D,
[f

(z)[
1 [f(z)[
2
1 [z[
2
.
4. Suppose f is an entire function, f is bounded for Re z 0, and f

is
bounded for Re z 0. Prove that f is constant.
5. (a) Suppose R(z) is a rational function with no poles on the unit circle.
Prove that
_
|z|=1
R(z) dz =
_
|w|=1
R
_
1
w
_
dw
w
2
.
(b) Use part (a) to evaluate the integral
_
|z|=1
z
11
12z
12
4z
9
+2z
6
4z
3
+1
dz.
6. Suppose g
n
is a sequence of non-constant entire functions. If the func-
tions g
n
converge uniformly to a function g, what can you conclude about
this sequence? When is the limit entire?
In the following parts, assume g
n
, n N, is a sequence of entire functions,
having only real zeros. Further, suppose that the functions g
n
converge
uniformly on compact subsets of C to an entire function g.
(a) Prove that if g is not identically zero, then g has only real zeros.
(b) It is possible for g to be identically zero? Explain your answer.
198 August 2004
11.7 August 2004
1. Let f be holomorphic on C 0. Suppose that for any z ,= 0,
[f(z)[ [log([z[)[ .
Show that f(z) 0.
2. Let C
1
and C
2
be two Euclidean circles in the plane with C
2
lying in the
interior of C
1
. Let be the domain bounded by these circles. Is there a
conformal map of bijectively onto an annulus z : 0 < r
1
< [z[ < 1? If
there is one, describe it; if there is not, describe why none can exist.
3. Prove or give a counterexample. Suppose f is holomorphic in D and
continuous on its closure. Then f extends to a holomorphic function on
B(0, R) for some R > 1.
4. Evaluate and justify your answer.
(a)
_

cos x
x
2
+ 1
dx.
(b)
_
2
0
d
(a +b cos )
2
, a > b > 0.
5. Let T = f : D D : f Hol(D), and L = sup
fF
[f

(0)[.
(a) Show that L exists (as a nite number).
(b) Show that there is a function f T such that f

(0) = L.
6. (a) Suppose is a bounded region in C, f Hol(), f ,= 0 in , f is
continuous on the closure of , and [f[ is constant on the boundary
of . Prove that f is constant on .
(b) Can the hypothesis that f ,= 0 in be dropped?
(c) Can the hypothesis that is bounded be replaced by the assumption
that the complement to is unbounded?
Preliminary Exams 199
11.8 August 2003
1. (a) Show that there is a complex dierentiable function dened on the
set = z C : [z[ > 4, whose derivative is
z
(z 1)(z 2)(z 3)
.
(b) Is there is a complex dierentiable function on whose derivative is
z
2
(z 1)(z 2)(z 3)
?
2. Evaluate the integral
_
2
0
e
e
i
d.
3. Suppose f and g are entire functions and [f(z)[ [g(z)[ for all z C.
Prove that there exists a constant c such that f = cg.
4. (a) Prove that every one-to-one conformal mapping of
D = z C : [z[ 1
onto itself is a linear fractional (Mobius) transformation.
(b) Prove that every one-to-one conformal mapping of D onto a disk
B = az C : [z a[ r for some a C and r > 0 is a linear
fractional transformation.
5. Show that f(z) = z/(e
z
1) has a removeable singularity at z = 0 and
that f has a power series expansion f(z) =

n1
c
n
z
n
. Calculate c
0
and
c
1
and show that c
2n+1
= 0 for n 1 (i.e., f is an even function). Find
the radius of convergence of the series.
6. Let 0 < r < R and A = z C : r [z[ R. Show that there exists a
positive number > 0 such that for each polynomial p,
sup
zA

p(z)
1
z

.
7. Evaluate the following real integral by using residues:
_

0
cos x
1 +x
2
dx =
1
2
_

0
_
e
ix
1 +x
2
+
e
ix
1 +x
2
_
dx.
200 August 2002
11.9 August 2002
1. Suppose f is holomorphic in a region that contains the closed unit disk
and [f(z)[ < 1 when [z[ = 1. How many xed points must f have in the
open unit disk D?
2. Suppose f is an entire function and there are constants A and B and a
positive integer k so that [f(z)[ A+B[z[
k
for all z. Prove that f must
be a polynomial.
3. Compute (justifying your computations):
(a)
_

x
2
1 +x
4
dx. (b)
_
2
0
d
a +b sin
, a > b > 0.
4. Suppose f is holomorphic and non-zero in the simply connected domain .
(a) If n is any positive integer, prove that there exists a function g,
holomorphic in satisfying g
n
= f.
(b) How many holomorphic solutions does g
3
= f have in a small disk
about 0 if f(z) := z
4
+ 16?
(c) Find the Taylor polynomial of degree 5 for the holomorphic solution
g in part (b) for which g(0) R.
5. Suppose is a region in C and Hol() denotes the space of functions
which are holomorphic in . Let f
n
be a locally bounded sequence in
Hol() and f Hol(). Assume A := z : limf
n
(z) = f(z) has
a limit point in . Show that there exists a subsequence of f
n
which
converges to f uniformly on compact subsets of .
6. In a domain containing 0, a function f : C dened by f(x, y) =
u(x, y) +iv(x, y) is complex harmonic if both u and v are (real) harmonic
in . You may assume that f admits an absolutely convergent double
power series expansion f(z, z) =

n,m0
a
nm
z
n
z
m
and that the usual
dierentiation and integration rules for power series in one variable are
valid here.
(a) Under what conditions on the coecients a
nm
is f holomorphic in ?
(b) Under what conditions on the coecients a
nm
is f complex harmonic
in ?
Preliminary Exams 201
11.10 August 2001
1. Show that the function u(z) := log [z[ has no harmonic conjugate in the
domain = C 0.
2. Evaluate
_
|z|=2
e
z
z
3
+z
dz.
3. Suppose f is analytic in the unit disk D and f(0) = f

(0) = 0. Also
assume that [f

(z)[ 1 for all z D. Prove that


[f(z)[
[z[
2
2
everywhere in D.
4. Show that all the zeros of
p(z) := 3z
3
2z
2
+ 2iz 8
lie in the annulus z C : 1 < [z[ < 2.
5. is dened as the intersection z C : [z[ < 1 z C : [z
1
2
[ >
1
2
.
Find a conformal map f of the region onto the open unit disk D. Can
you extend the mapping f to the closure of ? Is it conformal on part or
all of the boundary?
6. A continuous mapping f : X Y is called proper if, for any compact set
K Y , f
1
(K) is compact in X. Prove that f : C C is a polynomial
if and only if it is a proper, entire function.
7. Let D be the open unit disk and B be the open unit ball in C
2
; i.e.,
B := (z, w) : [z[
2
+[w[
2
< 1 C
2
.
Show that there cannot be any non-zero holomorphic function f, de-
ned on D, whose graph is contained in B. (Hint: What happens when
[z[ 1?)
202 August 2000
11.11 August 2000
1. Compute
_

cos x
(x
2
+ 1)
2
dx.
2. Let H denote the right halfplane; i.e., H := z : Re z 0. Given that
f : H H is holomorphic and f(1) = 1, show
(a) [f

(1)[ 1, and
(b)
[f(z) 1[
[f(z) + 1[

[z 1[
[z + 1[
.
3. Let f and g be entire functions with [f(z)[ [g(z)[ for all z C. Prove
that there exists a constant K so that f(z) = Kg(z).
4. Determine the number of zeros of the function g(z) = e
z1
az inside the
unit circle z C : [z[ < 1 assuming [a[ > 1.
5. Let Hol() be the set of functions holomorphic in a domain and suppose
that T Hol() is some normal family in . Prove that
T

:= f

: f T
is also a normal family.
6. Suppose that f is entire and f(z) is real if and only if z is real. Show that
f can have at most one zero in C.
7. Suppose that D is the unit disk and f is a holomorphic map of D into
itself with f(0) = 0. If
f
n
:= f f f
. .
n
,
state the conditions under which limf
n
exists in all of D. When the limit
does exist, what is it?
BIBLIOGRAPHY
[Ahl79] Lars V. Ahlfors, Complex analysis, third ed., International Series in
Pure and Applied Mathematics, McGraw-Hill, New York, 1979.
[Car95] Henri Cartan, Elementary theory of analytic functions of one and sev-
eral complex variables, Dover Publications, New York, 1995.
[CF] Sergio Santa Cruz and Paulo Soares Fonseca, Bombelli: A Java com-
plex function viewer, Last update: 06/10/2000.
[Con73] John B. Conway, Functions of one complex variable, I, Graduate Texts
in Mathematics, vol. 11, Springer-Verlag, New York, 1973.
[Fis90] Stephen D. Fisher, Complex variables, second ed., Dover, Mineola, NY,
1990.
[Hil59] Einar Hille, Analytic function theory: Volumes I & II, Ginn and Com-
pany, 1959.
[Kno52] Konrad Knopp, Elements of the theory of functions: Volumes I & II,
Dover Publications, 1952.
[Lan97] Serge Lang, Undergraduate analysis, second ed., Undergraduate Texts
in Mathematics, Springer-Verlag, New York, 1997.
[Rem91] Reinhold Remmert, Theory of complex functions, Graduate Texts in
Mathematics: Readings in Mathematics, vol. 122, Springer, New York,
1991, Translated from German by R. Burckel.
[Rud87] Walter Rudin, Real & complex analysis, third ed., Tata McGraw-Hill,
New Delhi, 1987.
[SL99] Rami Shakarchi and Serge Lang, Problems and solutions for Complex
Analysis, Springer-Verlag, New York, 1999.
203
INDEX
A(a, r, R), 118
arc, 33
dierentiable, 33
Jordan, 33
piecewise dierentiable, 33
argument, 4
Arzela-Ascoli theorem, 126

C, 5
Casorati-Weiestrass theorem, 82
Cauchys estimates, 75
Cauchys theorem
for derivatives, 75
for disks, 66
for rectangles, 64
general disk version, 67
Cauchy-Riemann equations, 11
chain rule, 14
closed curve, 33
complex conjugate, 4
component
connected, 31
conformal mapping, 35
continuous function, 10
contractable curve, 88
cosine, 25
cross ratio, 39
curve
closed, 33
dierentiable, 33
Jordan, 33
regular, 33
derivative, 10
dierentiable
arc, 33
curve, 33
directional derivative, 106
equicontinuous, 125
essential singularity, 82
exponential, 24
fundamental theorem of algebra, 76
Gauss-Green theorem, 107
Greens theorem, 59
for rectangles, 62
H, 37
harmonic, 103
harmonic conjugate, 13, 104
harmonic function, 13
homologous to zero, 87
homotopic, 87
Hurwitz theorem, 114
imaginary part, 4
index of a point, 71
integral, 57
isolated singularity, 80, 81
Jordan
204
INDEX 205
arc, 33
Jordan curve, 33
keyhole, 70
Laplacian, 13
Laurent series, 120
limit
of a function, 9
linear fractional transformation, 37
Liouvilles theorem, 76
logarithm, 26
Mobius group, 37
Mobius transformation, 37
maximum principle, 85
maximum principle for harmonic func-
tions, 109
mean value property for harmonic func-
tions, 109
meromorphic, 93
modulus, 4
Moreras theorem, 76
normal family, 124
order
of a pole, 17
of a zero, 16
path, 33
path integral, 57
piecewise dierentiable arc, 33
Poisson kernel, 112
pole, 17, 81
at , 17
power series, 19
power series expansion
cosine, 25
exponential, 24
sine, 25
precompact, 124
prelim
study guide, iii
syllabus, 129
primitive root, 4
principal part, 18
real part, 4
regular
curve, 33
dierentiable arc, 33
relative topology, 29
removable singularity, 77
residue, 92
residue computation at pole, 92
residue theorem, 93
reverse triangle inequality, 5
Riemann mapping theorem, 43
Riemann sphere, 6
Rouches theorem, 93
Schwarz lemma, 85
separation, 29
simply connected, 86
sine, 25
singular part, 18
spherical metric, 7
stereographic projection, 6
Taylors theorem, 79
theorem of
Abel, 23
Arzela-Ascoli, 126
Casorati-Weierstrass, 82
Cauchy, 64, 66, 75
Gauss-Green, 107
Green, 59
Hurwitz, 114
Liouville, 76
Lucas, 16
Morera, 76
Rouche, 93
Schwarz, 85
Taylor, 79
Weierstrass, 114
triangle inequality
proof, 5
reverse, 5
winding number, 71
zero, 16